**Multi-Hazard Risk Assessment at Community Level Integrating Local and Scientific Knowledge in the Hodh Chargui, Mauritania**

### **Maurizio Tiepolo 1,\* , Maurizio Bacci 1,2, Sarah Braccio <sup>1</sup> and Stefano Bechis <sup>1</sup>**


Received: 26 July 2019; Accepted: 9 September 2019; Published: 16 September 2019

**Abstract:** Hydro-climatic risk assessments at the regional scale are of little use in the risk treatment decision-making process when they are only based on local or scientific knowledge and when they deal with a single risk at a time. Local and scientific knowledge can be combined in a multi-hazard risk assessment to contribute to sustainable rural development. The aim of this article was to develop a multi-hazard risk assessment at the regional scale which classifies communities according to the risk level, proposes risk treatment actions, and can be replicated in the agropastoral, semi-arid Tropics. The level of multi-hazard risk of 13 communities of Hodh Chargui (Mauritania) exposed to meteorological, hydrological, and agricultural drought, as well as heavy precipitations, was ascertained with an index composed of 48 indicators representing hazard, exposure, vulnerability, and adaptive capacity. Community meetings and visits to exposed items enabled specific indicators to be identified. Scientific knowledge was used to determine the hazard with Climate Hazards Group Infra-Red Precipitation with Station (CHIRPS) and Tropical Rainfall Measuring Mission (TRMM) datasets, Landsat images, and the method used to rank the communities. The northern communities are at greater risk of agricultural drought and those at the foot of the uplands are more at risk of heavy rains and consequent flash floods. The assessment proposes 12 types of actions to treat the risk in the six communities with severe and high multi-hazard risk.

**Keywords:** agricultural drought; climate change; heavy rains; hydrological drought; meteorological drought; risk assessment; Sahel; sustainable rural development

#### **1. Introduction**

In the semi-arid rural areas of tropical Africa, drought reduces access to water for human consumption and affects both livestock and rain-fed crops. In the same areas, more and more frequent flash floods destroy irrigated crops, damage hydraulic works, and consequently prevent recession agriculture with which smallholder farmers compensate the deficit of rain-fed crops [1]. The coexistence of different hazards is so frequent that the Sendai framework for Disaster Risk Reduction (2015) recommends a sustainable livelihood development based upon multi-hazard risk assessments [2]. However, the risk assessments published on tropical Africa (Table 1) deal with a single hazard at a time, which in 90% of cases, is the flood hazard and, for the remainder, the drought hazard.



The published assessments usually follow two approaches. The first approach performs distance assessments through satellite images, digital elevation models, soil maps, precipitations, and stream-flow datasets. When the approach involves using indicators, it obtains them from literature rather than from ground truth. It neglects the risk reduction actions in progress and proposes few in case of drought. The second approach listens to the community. This is not sufficient to appreciate the risk of events that the community has not yet experienced [28], to assess the probability of disaster (which is a constitutive element of the concept of risk), or to identify the exposed areas.

Although the importance of local knowledge in adapting to climate change [29] is unquestionable and there is a broad consensus regarding the need to combine it with scientific knowledge in risk reduction [30–33], assessments in tropical Africa which integrate both types of knowledge are still rare [1,3,18].

The Sendai framework recommends improving the understanding of risks at all spatial scales [10]. Indeed, after Sendai, the subnational assessments have increased. The regional scale is important for action, as it is that in which official development aid and the local authorities themselves operate most often with medium-term programs and projects. However, regional scale risk assessments struggle to estimate the hazard when climate information is scarce [34], to identify truly expressive indicators of the risk determinants, to take account of the risk reduction measures in place, and to suggest pertinent ones for the future (Table 1).

At the regional scale, assessments which identify the most exposed communities and suggest how to deal with the risk could help official development aid and the local authorities in risk-informed decision-making, which would certainly aid sustainable livelihood development.

The aim of this article was to develop a multi-hazard risk assessment ranking the communities according to the risk level and identify risk reduction actions using local and scientific knowledge and techniques adapted to unskilled operators and scant information.

To achieve this objective, the multi-hazard risk index (MHRI) was proposed. The risk equation used combines hazard (H), meaning the "potential occurrence of a natural physical event that may cause loss of life, injury or damage to property" [35]; exposure (E), or "the presence of people, livelihoods, species or ecosystems, environmental functions, services, and resources, infrastructure, or economic, social, or cultural assets in places and settings that could be adversely affected" [35]; vulnerability (V), "the propensity or predisposition to be adversely affected" [35]; and adaptive capacity (AC), namely "the ability of systems, institutions, humans and other organisms to adjust to potential damage, to take advantage of opportunities, or to respond to consequences" [35]: R = H × (E + V − AC). The equation is an adaptation of that proposed by Crichton [36]. Each risk determinant is expressed by indicators, which were identified after discussion with the communities and visits to the exposed items.

The multi-hazard risk assessment was carried out with 13 rural communities of the municipalities of Adel Bagrou, Agoueinit, Bougadoum, Oum Avnadech, in Hodh Chargui, Mauritania, a landlocked region located 1100 km from the Atlantic Coast (Figure 1).

**Figure 1.** The 13 rural communities of Hodh Chargui where the multi-hazard risk assessment was developed.

Location names in local documents change. This article used the names adopted by the general census of the population of the Islamic Republic of Mauritania in 2013 [37]. These communities were among the most in need of reducing the hydro-climatic risk, according a preliminary survey on 271 communities (2017) of the Hodh Chargui. The 13 communities can be aggregated into three clusters: One around Néma, the capital of Hodh Chargui; one to the south-west of Néma; and one to the south, around Adel Bagrou, close to border with Mali. The Hodh Chargui is experiencing strong demographic growth: The region increased from 212,000 inhabitants in 1988 to 431,000 inhabitants in 2013 [37].

The 13 agropastoral communities have between 400 and 2600 inhabitants. These are settlements that have been established mostly in the last 40 years, constituted by the agglomeration of stone dwellings (sometimes made from crude earth bricks). Each stone dwelling is flanked by a construction with a two-pitched roof, under which life takes place during the warm months, and an enclosure for the animals around a tree in the branches of which the fodder is stored.

The key elements of each community are the wells which are, in general, traditional (uncovered, without water pumps), and are, in some cases, flooded during the wet season. In the dry season, they may be found some kilometers away. It is rare for a community to have a borehole with a respective water reservoir. The mosque, the school, and the community leader's residence are the other significant locations of each community. During the dry season (November–May), the shepherds go to the southern pastures. Only a small part of the livestock remains in the communities for requirements of milk, cheese, and meat. To sell the animals, the shepherds go as far as Senegal, traveling between 100 and 300 km [38].

Between June and August, rain-fed agriculture is practiced. The herds return to the pastures around the villages of origin. Half of the communities use an ephemeral wetland, at the edges of which they dig wells for pastoral and sometimes human use. In this season, some communities remain isolated because the conditions of the tracks do not allow vehicular access. Poor rainfall leads communities to pick up the runoff with earth embankments to practice recession agriculture from October onward. The embankments are, however, exposed to trampling by the herds which cross them or by the heavy rains. When the embankments are damaged, they no longer retain water and recession agriculture is reduced to small surfaces. The households that can afford it enclose the fields with metal fences to protect the crops from stray livestock. Those who do not have such resources resort to the use of thorny branches, which is prohibited, to preserve the scarce arboreal and shrubby coverage of the region. Later, when the dry season takes hold, irrigated commercial agriculture commences.

These activities are constantly exposed to drought, which manifests primarily with "an abnormal precipitation deficit" [35] (meteorological drought). This has effects that continue well beyond the wet season: The availability of fodder (trees, shrubs, grass) drops, the ephemeral wetlands where livestock are watered do not fill up, the surface aquifer does not refill, the pastoral wells and traditional wells used by many communities dry up or have poor quality water, and the possibility of recession agriculture and irrigated gardening fades.

Drought can also manifest with a "shortage of precipitation during the runoff and percolation season primarily affecting water surfaces" (hydrological drought) and with a "shortage of precipitation during the growing season" (agricultural drought) [35].

The three forms of drought co-exist with heavy precipitations, which damage the earth embankments, as well as make the wells in the flood areas inaccessible and, if flooded, unusable. All these events threaten the livelihood of the Hodh Chargui communities (Figure 2).

**Figure 2.** Main hazards and their impacts on livelihoods in the Hodh Chargui, Mauritania.

This assessment thus considered meteorological, hydrological, and agricultural drought and heavy precipitations. The four hazards are combined in the multi-hazard risk index (MHRI), respecting the same importance of each risk determinant and the quantitative measurement of the indicators [39].

The assessment is organized into four phases (Figure 3).

The first phase identifies the context: The trend of daily precipitation, the variation in the extension of surface waters, the risk criteria (equation, probability of occurrence, level), and the technique to be used. The second phase identifies the risk: Which datasets to use to determine the hazard and how to ascertain exposure, vulnerability, and adaptive capacity. The third phase identifies the indicators for each risk determinant, collects information from the 13 communities, processes the data, produces the MHRI, and represents it on the map. The final phase identifies the risk reduction actions.

The most significant results obtained by the assessment are the reproducibility of the methodology and the existence of very different risk levels in an apparently homogeneous territory, influenced, above all, by meteorological drought and heavy rains, and the proposal of 12 risk reduction actions.

**Figure 3.** Multi-hazard risk assessment flowchart.

#### **2. Materials and Methods**

The multi-hazard risk was ascertained using the index technique [40]. This technique fits unskilled operators and can be used in other regions of Mauritania.

The MHRI adds up the risk indices of the three types of drought and heavy precipitations. The index is made up of 48 indicators that quantitatively scored hazard, exposure, vulnerability, and adaptive capacity. The actual values found for each indicator were normalized within a 0–1 scale. The indicators were then added and normalized in a 0–1 scale for exposure, vulnerability, and adaptive capacity (Supplementary file 1). Twenty-nine vulnerability and exposure indicators were acquired through a survey in each community (April 2017), fifteen exposure and adaptive capacity indicators were measured during a visit at the end of the dry season (May 2018), and four hazard indicators were acquired from datasets on the daily and three-hourly rainfall and from satellite images (Figure 4). The survey (April 2017) and the subsequent visit (May 2018) encountered between 12 and 20 inhabitants in each community gathered in two separate groups: Herders (men), and horticulturalists, goat and sheep breeders, and sun-dried tomatoes, milk, yogurt, cheese, dried meat, leather, livestock feed sellers (mainly women). Each group was asked a series of questions (Supplementary file 2).

**Figure 4.** Indicators used in the multi-hazard risk index for 13 communities in the Hodh Chargui, Mauritania.

The meteorological drought hazard is expressed by the probability of occurrence of rainfall accumulation during the months of July, August, and September of less than 150 mm. That limit was identified based upon the quantity of rain considered the minimum amount necessary to produce plant biomass in an arid Sahelian environment [41–44]. Rainfall distribution is derived from the Climate Hazards Group Infra-Red Precipitation with Station (CHIRPS) dataset in the 1981–2018 period. Using these values, we calculated the annual rainfall accumulation and evaluated it on the 38-year historical series (1981–2018) to determine how many years there was a rainfall of less than 150 mm for each community.

The hydrological drought hazard was calculated on six ephemeral wetlands of reference for most of the communities considered. Its determination, due to the absence of localized information, did not use the current indices of hydrological drought [45]. The analysis was based upon the extension of the ephemeral wetlands as identified by calculating the Normalized Difference Water Index (NDWI) [46] on the Landsat satellite images. The availability of a limited series of images led us to identify the rainfall accumulation, which determined the years in which the surfaces of the ephemeral wetlands were less extensive. Then, we searched for the frequency of that value in the 1981–2018 rainfall series.

First, the annual surface profile of each ephemeral wetland in a dry year (2014) and in a wet year (2015) was determined. The satellite images were taken on different dates (*t*) each year, requiring the construction of the profile by interpolating the surface data with the formula:

$$Surface\_t = Surface\_{t-1} + \left(\frac{Surface\_{t+1} - Surface\_{t-1}}{N.Days\_{[(t+1)-(t-1)]}}\right) \times N.Days\_{[(t)-(t-1)]}$$

The results allowed us to construct the average surface area growth values (Figure 5).

**Figure 5.** Interpolated profile of Agoueinit and Vani ephemeral wetlands area in 2014 and 2015 and the average growth curve.

Using those average values, it was possible to calculate a standardized profile of the daily step curves for the July–November period. The standardization was calculated daily, using the maximum value recorded for each ephemeral wetland as a reference (Figure 6).

**Figure 6.** Normalized growth curves of the six ephemeral wetlands.

With only the 2014 and 2015 years available, we found a rough curve of variation of the surfaces of the ephemeral wetlands. However, with this method, it was possible to discriminate between the various ephemeral wetlands.

Using the CHIRPS estimated rainfall dataset, the rain signal was decomposed at various time intervals from the accumulation of 7, 14, 30, and 60 days before the measurement of the surface of the ephemeral wetland in question, the accumulation of the two central months of the rainy season (August–September), and the entire season. With those values available, we sought to understand which rainfall time interval most influenced the filling of each individual ephemeral wetland, and then analyzed the correlation. The results show a different sensitivity of each individual ephemeral wetland to the August–September precipitations (Table 2).


**Table 2.** Correlation between the precipitation-ephemeral wetland surface.

The period of rainfall that, on average, most influenced the filling of the ephemeral wetland was the accumulation of rain in the months of August and September. That time interval was used to identify the rainfall that characterized the three years with less surface of the ephemeral wetland. The average value of these three years was used to identify the critical rainfall and, consequently, the probability of occurrence on the entire series (1981–2018).

The adopted method remains more suitable to measure changes in water bodies over time than that proposed by the European Commission Joint Research Centre, which reports the status of the individual pixels that make up the water bodies without, however, reporting the precise date of which they are observed [47].

The agricultural drought hazard was measured with the probability of occurrence of dry spells of at least 10 consecutive days during the months of July, August, and September, ascertained using the CHIRPS dataset for the period 1981–2018 for each community. The choice of this spell length reflected the need to consider a threshold that could generate a negative impact on rain-fed crops in this region, which is strictly dependent on the specific crops' resistance to drought stress [48].

The heavy precipitation hazard is expressed by the probability of occurrence of three-hourly rainfalls higher than 20 mm ascertained using the Tropical Rainfall Measuring Mission (TRMM) dataset for the 1991–2014 period at each ephemeral wetland.

The exposure to meteorological drought is given by the presence of irrigated crops, the number of inhabitants per well, and tropical livestock units [49] which remain in each community in the dry season: The higher the number, the more the community is exposed to drought. The exposure to hydrological drought is represented by the number of ponds, earth dams, and inhabitants of each community. As for exposure to agricultural drought, this is given by the presence of horticultural activities protected by barbed wire fencing against the intrusion of stray cattle and the presence of pasture and arable surfaces. A proxy indicator is the share of bare land in the territory of each community: The lower it is, the greater the exposure. The exposure to heavy precipitations is given by the number of earth embankments and by the wells and houses in flood prone areas, which could become inaccessible or be flooded.

The vulnerability to meteorological drought is given by the distance of the pastures from the village in the dry season. The vulnerability to hydrological drought is expressed by distant wells with poor water flow and quality and by the lack of boreholes, functioning fountains, or by broken diesel water pumps. It is also expressed by the population growth rate of the community in question, which increases the demand for water. The vulnerability to agricultural drought is still linked to the availability and accessibility of wells for irrigation, the practice of cropping for self-consumption only, the rate of unfenced lots, the distance to the market, and the number of days of road interruption. The vulnerability to heavy precipitations is expressed by the possibility of receiving an early warning by telephone (therefore, the coverage of the area with a mobile phone signal), by the lack of protection of the earth embankments from the crossing of livestock, by the absence of spillways and locks, which reduce the pressure of flash floods on the hydraulic works, and by the presence of creeks without bank protection.

The adaptive capacity is of three types [50]: Capacity to anticipate risk, to respond to risk, and to recover and to change. For meteorological drought, the existence of radio programs aimed at farmers who report where vaccines and vaccination parks for livestock (anticipate), pastures, water, and fodder banks (recover) are available is paramount. For hydrological drought, the existence of boreholes, fountains or mini aqueducts (respond) which cover the demand for water by drawing from deep aquifers, especially if powered by solar water pumps, which have lower operating costs than diesel water pumps, is important.

Agricultural drought can be overcome if there are extension services (anticipate) and farmers' associations (recover). For heavy precipitations, radio access counts to receive early warning (anticipate). Spillways and locks in the earth embankments allow the pressure of flash floods on the earth embankments to be regulated, preserving them from collapse (respond).

In order to compare the different hazards, we took a series of precautions [39]. Each indicator and each determinant has the same significance. The probability of occurrence of each hydro-climatic hazard was calculated observing the same timeframe (1981–2018), except for heavy precipitations (1998–2014).

For each individual risk, the value of every individual determinant ranges between 0 and 1, irrespective of the number of indicators that describe it. Each indicator has the same significance. The MHRI was obtained by adding the values of the four single risks. Its value can theoretically vary from 0 to 8. The absolute interval between the maximum and minimum value was divided into four equal parts to indicate low (0–2), moderate (2–4), high (4–6), and severe (6–8) risk. In reality, the highest value can never be reached because the hazard is always less than 1 and all the communities have risk reduction actions in place that bring the value of the adaptive capacity higher than 0.

The calculation allowed to identify which risk, determinant, and indicators have the greatest effect on the MHRI. The indicators that present the highest value (exposure, vulnerability) or lowest value (adaptive capacity) oriented the identification of risk treatment actions among the best practices developed in the Hodh Chargui region. Each of these were then assessed with the communities, considering the expected impact on sustainable rural development, the successful use in the region, the community participation in construction works, the maintenance requirements, the maintenance local capacities, and the community acceptance. One point was attributed to each criteria. The resulting ranking identified priority actions.

#### **3. Results**

#### *3.1. Hazard*

#### 3.1.1. Meteorological Drought

It is somewhat challenging to define meteorological drought in Hodh Chargui. The scarcity of observed data does not allow us to understand the quantity of precipitation that can generate a negative influence on the production system. Furthermore, the production systems are naturally resistant to extreme drought conditions. Literature identified 100–150 mm of the annual rainfall as the threshold within which plant species that are most resistant to drought can produce biomass even in extreme conditions [41–44]. For this work, we decided to use the 150 mm threshold for meteorological drought. Using the CHIRPS series, the rain profile of the region was extracted. In the last 38 years, the Hodh Chargui had its driest period during the 1980s, with a minimum of 102 mm in 1983. In the last 13 years, there has been a trend toward the recovery of rainfall, with values that never dropped below 150 mm per year commencing from 2006 (Figure 7).

**Figure 7.** Yearly precipitation 1981–2018 in the southern Hodh Chargui region by Climate Hazards Infra-Red Precipitation with Station (CHIRPS) dataset and 150 mm limit.

The rainfall distribution follows a south-north gradient starting from 300 to 150 mm/year. Nevertheless, during the 2010–2018 period, there was a recover in rainfall of 30–50 mm/year throughout the region (Figure 8).

**Figure 8.** The difference in average accumulated rainfall during 2010–2018 and 1981–2010.

Boukahzama 1, Agoueinit, and NGuiya are the northernmost communities and are most likely to have rainfall of less than 150 mm. Conversely, the five communities on the border with Mali (Drougal, Gnebett Ehel Heiba, Jrana, Mberey El Jedida, and Goubya Elmesjid) have a very low probability of meteorological drought (Table 3).


**Table 3.** Meteorological drought probability for 13 communities of the Hodh Chargui.

#### 3.1.2. Hydrological Drought

Six ephemeral wetlands constitute the surface water resources of reference for the investigated communities. These are semi-permanent and shallow water bodies of maximum extension between 6 and 30 km2. All are characterized by weak depth. Despite this, they are a fundamental resource for human and pastoral water supplies, for fishery resources, and in the case of Agoueinit, for recession agriculture. The flood regime is not the same. It is rare for a dry or wet year to affect all six ephemeral wetlands. In 2003, the ephemeral wetlands reached, in total, 78% of the maximum surface, followed by 2011 (70%), 2009 (57%), and 2012 (55%). In 1987, five ephemeral wetlands out of six had a surface reduced to less than 10% of the maximum observed extension. In 2005, it was so dry that the average surface of the ephemeral wetlands dropped to 1% of the maximum average. In 2014, a hydrological drought was experienced by half of the ephemeral wetlands, while two out of six experienced a hydrological drought in 2016. The two southernmost ephemeral wetlands have not suffered drought in the last eight years, while the northernmost has been dry for six years out of eight, and the center-south has been dry for one or two years out of eight (Figure 9).

**Figure 9.** Minimal (**left**) and maximal (**right**) surface of six ephemeral wetlands in the Hodh Chargui, Mauritania, 2001–2018.

According to the described methodology, which attributes the extension of six ephemeral wetlands to rainfall, it is possible to establish the drought probability on the same period used to estimate the meteorological drought probability. It follows that the ephemeral wetland of Agoueinit is particularly prone to drying out compared to all the others (Table 4).

**Table 4.** Hydrological drought probability for the six ephemeral wetlands in the Hodh Chargui.


#### 3.1.3. Agricultural Drought

Agricultural drought occurs when drought affects agricultural and pastoral production. The 13 communities in question cultivate in rain-fed conditions, in the form of recession agriculture, and of irrigated gardens. The first type of agriculture is influenced by the rainfall distribution. Pastoral production, which plays a major role in the region's economy, is influenced by the presence of pastures for transhumant herds during the wet season. Biomass production is therefore relegated to spontaneous herbaceous and shrub species, which are naturally very resistant to water stress conditions. However, lengthy dry spells during the wet season can reduce the availability of fodder. Using the CHIRPS dataset, the daily series of the different communities was extracted and the maximum length of the dry spell during the season was assessed (Figure 10).

**Figure 10.** Maximum dry spell length in July, August, and September in Hodh Chargui using the CHIRPS dataset.

From 2006 onward, rainfall is favorable, although it is always below 300 mm/year. However, there are three years with dry spells equal to or greater than ten days. The frequency of dry spells follows a north-south distribution. Dry spells are more frequent in the northern communities of Boukhzama 1, Agoueinit, and NGuiya, and are less frequent in the southern communities of Drougal, Gnebett Ehel Heiba, Goubya Elmesjid, Jrana, and Mberey El Jedida (Table 5).


**Table 5.** Agricultural drought probability expressed by dry spells frequency in 13 communities of Hodh Chargui, 1981–2018.

#### 3.1.4. Heavy Precipitations

It is difficult to define heavy precipitation in Hodh Chargui. The data are scarce and only one study has allowed for the definition of a threshold of extreme rainfall (37 mm/day) [51]. That value does not necessarily involve the generation of conditions favorable to flash floods. Those phenomena are linked to high rainfall intensity in periods of less than one day. As a consequence, the frequency of three-hourly rainfall higher than 20 mm was verified by analyzing the extractions of the three-hourly values from the Tropical Rainfall Measuring Mission (TRMM) dataset in the 1991–2014 period for each of the 13 communities. It follows that Boukhzama 1, Drougal, and Gnebett Ehel Heiba exceed this threshold more frequently than Elkenar, Jrana, and Mberey El Jedida. In this case, there is no decreasing distribution of the frequency of the hazard as it proceeds from north to south (Table 6).


**Table 6.** Heavy precipitations (>20 mm in 3 h) probability according the Tropical Rainfall Measuring Mission (TRMM) dataset, 1998–2014.

#### *3.2. Exposure*

The exposure to different hazards is represented by 11 indicators: Three for meteorological drought (irrigated crops, residential livestock, and number of inhabitants per well), three for hydrological drought (ponds, earth dams, and population), two for agricultural drought (bare land rate, fenced fields) and three for heavy rains (earth embankments, houses, and wells in flood prone area). The highest values of exposure to meteorological, hydrological, and agricultural drought and to heavy rains are reached respectively in Agoueinit, Vani, Agoueinit, and Boukhzama 1. The exposure to all hazards shows that Agoueinit and Legdur have the highest values and Mberey El Jedida has the lowest value.

#### *3.3. Vulnerability*

The vulnerability to the different hazards is represented by 23 indicators: One for meteorological drought (distance to pasture in dry season), eight for hydrological drought (electricity access, distance to wells, borehole not working, diesel water pump broken down, irregularly functioning fountain, wells water flow and quality, population growth rate), six for agricultural drought (wells access for gardening, absence of gardening due to lack of water, gardens fencing, distance to market, cropping for self-consumption, road interruptions), eight for heavy rains (mobile telephone signal and use, earth embankment absence, leaking, lacking spillway, fence, wells flooded, unprotected creek banks). The highest values of vulnerability to meteorological drought are reached by Legaida, while those to hydrological drought are reached by Vani, those to agricultural drought by Goubya Elmejid, and those to heavy rains by Legdur. The vulnerability to all hazards shows that Legaida and Legdur have the highest values, while Agoueinit and Vani have the lowest values.

#### *3.4. Adaptive Capacity*

The adaptive capacity to the different hazards is represented by 10 indicators: Three for the meteorological drought (herders/farmers radio programs, extension services for herders, fodder stock), two for the hydrological drought (fountain and boreholes), two for the agricultural drought (solar water pumps and small household farmer's associations) and two for heavy rains (radio access, and earth embankments provided with spillway). The highest values of adaptation to meteorological drought are reached by Elkenar and NGuiya, those to hydrological drought by Agouenit, those of adaptive capacity to agricultural drought by Boukhzama 1, and those of adaptive capacity to heavy rains by Drougal, Elkenar, Gnebett Ehel Heiba, Goubya Elmesjid, Legaida, Legdur, Nguiya, and Vani. The adaptive capacity to all hazards is highest in Goubya Elmesjid and lowest in Agoueinit, Begou and Legaida.

### *3.5. Multi-Hazard Risk Level*

The interval between the maximum and minimum value of the multi-hazard risk index (MHRI) was split into four parts of the same breadth to represent the severe (0.83–1.09), high (0.56–0.82), moderate (0.28–0.55), and low risk (0–0.27). It follows that NGuiya, Agueinit, and Begou are at severe risk, Legdur, Boukhzama 1 and Legaida are at high risk, Gnebett Ehel Heiba, Jrana, and Elkenar are at moderate risk, and all other communities are at low risk (Figure 11, Table 7). Therefore, it can be said that the most northern communities tend to have the highest risk levels and the five southernmost communities tend to have a low to moderate risk level. The value of the risk index was substantially determined by that of agricultural drought and heavy rains (Table 7).

**Figure 11.** Thirteen rural communities at multi-hazard risk in the Hodh Chargui, Mauritania.


**Table 7.** Multi-hazard risk index for 13 communities of Hodh Chargui, Mauritania.

#### *3.6. The Use of the Multi-Hazard Risk Index for the Identification of Risk Treatment Actions*

The risk assessment was aimed at the official development aid active in the region and at the regional administration. However, the method may also be of interest for other contexts exposed to similar hazards. In addition to the ranking of the communities according to risk level, the assessment process proposes thirteen risk treatment actions for the six communities at severe and high risk. These actions were identified after a visit to the exposed items. This involves acting on the water supply (well deepening, building or raising the apron of the pastoral wells, covering them, equipping them with pedals or solar-powered water pumps, water troughs for cattle watering, guaranteeing access even in the wet season), protect crops from stray cattle (fencing in metal barbed weir), and reduce the impact of heavy rains (repairing the earth embankments and equipping them with spillways or repairing the existing spillway, locks and fencing in metal barbed wire, gabion wall to protect the riverbank) (Figure 12, Table 8).

**Figure 12.** Open well (**1**) showing low apron (**2**) and basement (**3**), water trough for cattle watering (**4**), solar panels (**5**) in Boukhzama 1; a well provided with raised basement (**6**), a manual water pump (**7**), and a broken water trough (**8**); an earth embankment (**9**) provided with spillway (**10**) and lock (**11**) in Agoueinit.


**Table 8.** Risk treatment for the five communities at severe and high risk of the Hodh Chargui, Mauritania.

These measures, compared with those proposed by the literature, are more specific and directly implementable (Table 9).

**Table 9.** Comparison of risk reduction actions identified for Hodh Chargui with those suggested by other assessments.


#### **4. Discussion**

This study reviewed the problems common to the majority of risk assessments published thus far in tropical Africa. The study then proposed a multi-hazard vision, the integration of local and scientific knowledge, a drought and flooding probability estimate, indicators representing the risk determinants, and 12 actions to deal with the risk.

The objective was to define a replicable methodology that would allow us to produce a ranking of the communities and to suggest actions for those that would have been at greatest risk. Therefore, it was necessary to refer to specific communities with their own hazards, exposures, vulnerabilities, and adaptive capacities, rather than representing communities as points of a risk map constructed by superposition of low-resolution information layers, as is often done in assessments at regional scale.

This holistic approach found little evidence in the literature consulted at the regional scale. Local knowledge and visits to the exposed items facilitated the determination of the exposure, vulnerability, and adaptive capacity.

Scientific knowledge was used to determine which of the four hazards pose the highest threat to the sustainable development of livelihoods in the rural Hodh Chargui and considered the method used to rank the communities. Contrary to the indications of the Sendai framework (2015) [2], the integration of knowledge is still unusual on a regional scale. The systematic review highlighted that only one assessment out of four published on tropical African regions estimated the probability of flooding or drought [1,3,11,18,19,27]. This is likely to result from poor access to local data. In the case of the Hodh Chargui, for example, the only surviving weather station with a continuous series of more than 30 years of daily precipitation data is that of the airport of Néma, which is too little to represent a vast territory such as that in which the 13 communities are distributed. The use of estimated precipitation from satellite (CHIRPS dataset) and the surface area of the ephemeral wetlands (Landsat images) allowed for the estimation of the probability of meteorological, hydrological, and agricultural drought, and the TRMM dataset allowed for the estimation of the probability of heavy precipitation. The low spatial resolution of the Landsat images used (30 m) was adequate for the size of water bodies observed (6 to 30 km2).

The literature review ascertained that in risk assessments at the regional scale, the indicators are chosen according to the information most easily accessible, rather than according to the information that best represents the risk determinants. In particular, it is rare to find specific indicators for meteorological, hydrological, and agricultural drought. Meeting with the communities and the visits to the exposed items allowed for indicators specific to the context to be identified.

The considerable differences in the level of agricultural drought and heavy rains risks among the 13 communities have generated a differentiated multi-hazard index. The northernmost communities were found to have greater probability of agricultural drought risk, compared to the five southern communities on the border with Mali. However, they are also closer to the large market of Néma (22,000 inhabitants in 2013,) which demands many horticultural products in a region in which they are scarce. Therefore, they would have greater opportunities to diversify their livelihood with commercial gardening if they could improve their access to water. The communities at the foot of the uplands (Boukhzama 1 and Begou) are more exposed to the risk of heavy rains and, therefore, to flash floods.

The discussions with the communities and the visit to the receptors enabled the identification of twelve actions for the six communities at severe and high risk, which are rarely found in literature [11]. These first concern the improvement of access to water: Well deepening, apron elevation, covering, providing a pedal or a solar water pump, a water trough for cattle watering, and facilitating access in flood prone area during the wet season. Second, they concern earth embankments (creation of spillways and locks, protection with metal barbed wire) and the protection of the riverbanks (gabion wall).

The assessment methodology may be of interest for other semi-arid, agropastoral regions of the Tropics.

The assessment presents four main limits. First, it was based on a review that has only considered the published literature. Grey literature contains other assessments, but its dissemination is ephemeral and it has a temporary inspiring effect on assessment practices. For these reasons, it was not taken into consideration.

Second, the hydrological hazard was calculated in a very simplified manner, seeking to achieve a result despite the scarcity of available information at local scale. The correlation between the dynamics of the surface area of ephemeral wetlands and precipitations in a Sahelian context has been questioned by literature, particularly due to the alteration of the vegetation cover and related erosive processes that have increased the runoff over time [52–54]. However, in the specific context of southern Hodh Chargui, the almost flat orography may have limited the erosive processes compared to other areas at the foot of the uplands as Boukhzama 1 and Begrou [54].

Third, the occurrence probability of heavy precipitations was calculated on a shorter series than the conventional one (24 years instead of 38) due to the time limitations of the dataset used.

Fourth, water flow and the quality of the wells were appreciated qualitatively. Measuring these parameters is possible and would give further solidity to the assessment. However, as the number of investigations increases, the duration of the survey is extended and may risk limiting the comparison between the first and the last investigated communities, which was instead the objective of the assessment.

#### **5. Conclusions**

The recommendation of the Sendai framework (2015) to develop risk knowledge on a subnational scale is having effects in tropical Africa. In the last four years, hydro-climatic risk assessments on a regional scale have been increasingly practiced. However, the holistic approach, the integration of local and scientific knowledge, the assessment of the hazard, and the recommendation of specific actions are rarely practiced. The objective of this article was to propose a multi-hazard risk assessment method replicable in other semi-arid contexts of the Tropics that adopted a holistic approach by integrating local and scientific knowledge, ranking the communities according to the risk level, and listing actions to reduce the risk.

Of the thirteen communities where the assessment was developed, six were found to be severe and high risk according the relative intervals (low risk according the absolute intervals). The level of multi-hazard risk varies significantly, as it is mainly influenced by the risk of heavy rains and agricultural drought. The proposed actions are consequently detailed and largely differ from the generic remedies proposed by the literature. They concern the improvement of water access for agropastoral and human use, and the protection of hydraulic works and riverbanks from flash floods.

The proposed method can be extended to other Hodh Chargui communities and can be replicated in semi-arid agropastoral regions of the Tropics. The methodology requires less time and lower costs compared to assessments conducted exclusively through community surveys and provides more precise and articulated results.

The assessment helps to prioritize the actions depending on the value of the MHRI and to identify them in accordance with the prevailing risk. The assessment can also allow decisionmakers to monitor and assess the changes in the risk level occurring over time.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2071-1050/11/18/5063/s1, Table S1: MHRI-Multi-Hazard Risk Index; Table S2: Survey questions.

**Author Contributions:** Conceptualization, M.T.; Methodology, M.T. and M.B.; Investigation, M.T., M.B., S.B. (Stefano Bechis) and S.B. (Sarah Braccio); Writing—original draft preparation, M.T. and M.B.; Writing—review and editing, M.T., M.B., M.B. and S.B. (Sarah Braccio); Visualization, S.B. (Sarah Braccio); Funding acquisition, M.T.

**Funding:** This research was funded by DIST-Politecnico and University of Turin.

**Acknowledgments:** We are grateful to the two anonymous reviewers who offered enormously helpful comments to an earlier draft. We are, of course, responsible for any errors. We would like to thank Laura Alunno (Terre Solidali), Yeslem Hamadi, Lebatt Moubark (RIMRAP project) for facilitating the field activities, Souleymanou Cheikh Shed Bouh (Mayor of Agoueinit municipality) and Mohamed Lemine ould Abeid (Office National de la Météorologie) for information.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Risk Aversion, Inequality and Economic Evaluation of Flood Damages: A Case Study in Ecuador**

#### **Vito Frontuto 1,\* , Silvana Dalmazzone 1, Francesco Salcuni <sup>1</sup> and Alessandro Pezzoli <sup>2</sup>**


Received: 14 October 2020; Accepted: 27 November 2020; Published: 2 December 2020

**Abstract:** While floods and other natural disasters affect hundreds of millions of people globally every year, a shared methodological approach on which to ground impact valuations is still missing. Standard Cost-Benefit Analyses typically evaluate damages by summing individuals' monetary equivalents, without taking into account income distribution and risk aversion. We propose an empirical application of alternative valuation approaches developed in recent literature, including equity weights and risk premium multipliers, to a case study in Ecuador. The results show that accounting for inequality may substantially alter the conclusions of a standard vulnerability approach, with important consequences for policy choices pertaining damage compensation and prioritization of intervention areas.

**Keywords:** natural disasters; flooding; flood vulnerability; inequality; risk premium; expected annual damages; certainty equivalent annual damages; equity weight expected annual damages; equity weight certainty equivalent annual damage

#### **1. Introduction**

Flooding, defined by the Intergovernmental Panel on Climate Change (IPCC) [1] as 'the overflowing of the normal confines of a stream or other body of water or the accumulation of water over areas that are not normally submerged', is one of the most common and destructive natural disasters. Estimates of both affected people and economic losses vary widely. According to the Organization for Economic Co-operation and Development (OECD) [2], floods affect up to 250 million people in the world every year. In 2019, floods caused over 5000 casualties worldwide [3].

Population growth is driving an increase in the number of people living in areas susceptible to flooding, with a consequent surge in impacts on lives, properties and productive assets. Urbanization and development reduce the water retention capacity of soils and increase runoff [4]. Climate change is increasing the frequency and intensity of flood disasters throughout the world, which nearly doubled in 2000–2009 compared to the previous decade [5]. This combination of demographic, development and climatic drivers challenges societal resilience to catastrophic flood events. New data released by the World Resource Institute in April 2020 forecast the number of people harmed by floods to double globally by 2030. According to the projections obtained in 2019 by the Aqueduct Floods modeling tool of the World Resource Institute [6], damages to urban property are expected to rise from USD 174 to USD 712 billion per year.

The structure of impacts is not uniform across the world: low-income countries suffer higher fatalities, whereas high-income countries register higher values of damage to properties and infrastructures. Low or lower-middle-income countries accounted for 49 percent of flood events

recorded in the International Disaster Database EM-DAT between 1971 and 2015 and for more than 60 percent of all deaths. High and upper-middle-income countries accounted for just under 80 percent of the monetary value of all reported material damages from flood events [2].

The socio-economic significance of the issue and the expectation of an escalating trend stimulated a vast and fast-growing literature on economic impacts of flooding, particularly in urban contexts. McClymont et al. [7] provide a thorough account of the literature on flood risk management and resilience. Hennighausen and Suter [8] explore the impact of flood risk perception in the housing market in the US. Shatkin [9] develops a conceptual framework for assessing the implications of flood risk for urban development, considering issues of property rights, informality, neoliberalization and financialization and the role of the state, with a particular focus on Asian megacities. Goh [10] explores the interrelationships between biophysical factors (ecological scales of the watershed) and socio-political factors (infrastructural scales associated with flood protection, social and spatial marginalization) behind urban flood risk, based on field research in Indonesia. Chen et al. [11] study flooding-migration relationships by combining nationally representative survey data with inundation measures derived from weather stations and satellites. Oosterhaven and Tobben [12] propose a method to estimate the indirect impacts of flood disasters and apply it to the major 2013 flooding event of southern and eastern Germany. Kashyap and Mahanta [13] provide an in-depth review of previous literature.

As both latitude and poverty play a major role in explaining exposure to natural disasters, a number of case studies have focused on developing regions: Ogie et al. [14] on coastal megacities of developing nations, Cobian Alvarez and Resosudarmo [15] on Indonesia, Reynaud et al. [16] on Vietnam, De Silva and Kawasaki [17] on Sri Lanka, Erman et al. [18] on Tanzania, Kurosaki [19] on Pakistan, to cite a few.

A number of studies have also examined the vulnerability and response of different socio-economic groups to natural disasters (e.g., Rasch [20]; Rodriguez-Oreggia et al. [21]; Glave et al. [22]; Lopez-Calva and Ortiz-Juarez [23]; Carter et al. [24]; Brouwer et al. [25]; Masozera et al. [26]) as well as the relationship between poverty and disasters (Tahira and Kawasaki [27]; Borgomeo et al. [28]; Henry et al. [29]; Patnaik and Narayanan [30]; Hallegatte et al. [31]).

There is however, in our view, a yet understudied area of enquiry—the one concerning the methodological aspects of the valuation of economic impacts. Monetary estimates of economic losses from flooding play a crucial role in informing decisions and setting priorities on risk mitigation investments as well as in determining post-disaster compensations. Yet, there are no generally agreed principles on which to ground impact valuations, which partly explains the very large variance across estimates provided even by the most authoritative sources. Particularly lacking, in our view, is a shared methodological approach to account for income inequality in determining the real welfare impact of natural disasters. Simply summing individuals' monetary equivalents is likely to provide a misleading picture of relative impacts and inappropriate policy implications when flooding disproportionately affects the poor, for whom even the loss of everything may amount to small absolute monetary values.

In fact, in standard Cost-Benefit Analyses (CBA), as commonly implemented by governments and international agencies, policies are typically evaluated by summing individuals' monetary equivalents without any distributional concern (e.g. The guidelines for CBA issued by the OECD [32], the European Commission [33], the U.S. Environmental Protection Agency [34]) The same considerations hold generally also for guidelines specific to flood damage assessments (e.g. [35,36]).)

The issue of using distributional weights in CBA dates back to the 1950s [37], but recent literature shows that this discussion has been largely ignored in real world practice (inter alia Drupp et al. [38] and Adler [39]). Kind et al. [40] have suitably tackled the issue and proposed a social welfare approach to CBA for flood and other disaster risk management, showing with a simulation how considering income distribution can lead to different conclusions 'on who to target, what to do, how much to invest and how to share risks' (p. 1). If confirmed, their results would enable decision makers to improve the effectiveness and equitability of flood management policies. However, their methodological approach has not yet been tested in real world studies.

The objective of our work is to contribute to fill this gap. After presenting the methodological options through which we can consider income distribution in the evaluation of flood damages, we offer an illustration based on empirical data from a region of high flood vulnerability and significant income inequality, the Duràn Canton in the Guayas province of Ecuador. The analysis confirms that accounting for inequality substantially alters the ranking of different areas in terms of vulnerability to flood damages and thus provides important insights for policy choices pertaining damage compensation and prioritization of intervention areas.

The paper is organized as follows. In Section 2, we formally describe the four alternative evaluation methodologies proposed in previous studies to estimate flood damages. In Section 3, we present the context of the case study and the data on which the analysis is based. Then we develop the empirical analysis, by calculating (in Section 4) the equity weights and the risk premium multipliers required for the inequality-adjusted evaluation of damages, the results of which are illustrated and discussed in Section 5. Section 6 concludes the paper.

#### **2. Evaluation Methodologies**

Following Kind et al. [40], we consider four different methodologies to estimate costs and benefits of flood risk reduction.

The first is the standard estimation of the Expected Annual Damage (EAD). Damages are derived from the stage-damage (or depth-damage) function, which provides estimates of the total damages due to a flood given its depth. Total damages are then divided by the probability of flooding (inverse of the return period). EAD focuses on damages to buildings and it does not take into account diminishing marginal utility of income or risk premia. It is the procedure generally used to evaluate damages in a standard CBA (for applications to flood risk assessment, see for example Skovgård et al. [41], Dupuits et al. [42], Alian et al. [43]). Even though it does not accurately reflect welfare economics theory, it may represent a satisfying proxy in situations where the institutional setting provides compensations for flood damages and the latter do not represent a major share of disposable incomes.

A first factor neglected in standard valuations of expected damages, as already discussed in Schulze and Kneese [44], is risk aversion. Risk-averse people, in order to protect themselves from adverse events, are willing to pay an amount larger than the expected damage (ED)—which is what makes insurance markets feasible. Additional Willingness to Pay (*WTP*) above the reduction of ED is the risk premium. We assume a typical [45] risk-averse utility function—a concave curve that becomes flatter as income increases—with constant elasticity:

$$\mathcal{U}(\mathbf{Y}) = \frac{\mathbf{Y}^{1-\mathbf{y}}}{1-\mathbf{y}} \tag{1}$$

where *Y* is income and γ is the elasticity of marginal utility of income—the variation of utility in response to changes in income. For this utility function we can express the risk premium multiplier (*RM*), following the European Commission's guidelines to CBA [33], as:

$$RM = \frac{WTP}{EAD} = \frac{1 - \left\{1 + P\left[\left(1 - Z\right)^{\left(1 - \gamma\right)} - 1\right]\right\}^{\frac{1}{\left(1 - \gamma\right)}}}{PZ} \tag{2}$$

where the numerator is the *WTP* for flood risk reduction, the denominator is the expected damage, *P* is the probability of flood occurrence (inverse of the return period) and Z is the share of income eroded by the flood—the commonly adopted measure of vulnerability. The multiplier increases more than proportionally with vulnerability.

One possible monetary evaluation approach accounting for risk aversion consists in evaluating costs and benefits of disaster prevention or remediation policies on the ground of a certainty equivalent, calculated by multiplying the expected damage by the risk premium multiplier defined above. The resulting measure, called by Kind et al. [40] Certainty Equivalent Annual Damages (CEAD), weighs *WTP* by a factor that increases more than proportionally with the fraction of household income lost, so as to account for the fact that economic theory and empirical evidence make us expect more

socio-economically vulnerable individuals to be more risk averse. When compensation programs are insufficient to cover actual damages and these damages may erode a significant portion of incomes, adopting CEAD in CBA is a useful improvement over EAD.

The two approaches above do not take into account that marginal disutility of losses may vary substantially with the income of affected households, as predicted by welfare economics (and estimated in over 50 countries by Layard et al. [46]). The limits of CBAs weighing all benefits and costs equally regardless to whom they accrue—an issue thoroughly discussed in theory, besides Adler [39], also by Fleurbaey and Abi-Rafeh [47], Anthoff et al. [48] and the UK Greenbook [49]—become increasingly relevant in contexts where compensation is negligible, socio-economic vulnerability is high and income distribution is strongly unequal.

Given a standard utilitarian welfare function *W* = *f*(*U*1, *U*2, ... , *UN*), a change in social welfare can be written as the sum of the marginal contribution to social welfare of the variation in utility of each individual:

$$
\partial \mathcal{W} = \left( \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_1} \partial \mathcal{U}\_1 + \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_2} \partial \mathcal{U}\_2 + \dots + \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_N} \partial \mathcal{U}\_N \right) \tag{3}
$$

If we consider a change in income:

$$
\partial \mathcal{W} = \left( \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_1} \frac{\partial \mathcal{U}\_1}{\partial \mathcal{Y}\_1} \partial \mathcal{U}\_1 + \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_2} \frac{\partial \mathcal{U}\_2}{\partial \mathcal{Y}\_2} \partial \mathcal{U}\_2 + \dots + \frac{\partial \mathcal{W}}{\partial \mathcal{U}\_N} \frac{\partial \mathcal{U}\_N}{\partial \mathcal{Y}\_N} \partial \mathcal{U}\_N \right) \tag{4}$$

Equity weights can be derived, as done, for example, in Fleurbaey and Abi-Rafeh [47] and the European Commission [33], by summing one monetary unit to a person's annual income and calculating the variation in utility:

$$
\partial \mathcal{W} = (\omega\_{\mathcal{U}\_1} \cdot \omega\_{\mathcal{Y}\_1} \cdot \partial \mathcal{Y}\_1 + \omega\_{\mathcal{U}\_2} \cdot \omega\_{\mathcal{Y}\_2} \cdot \partial \mathcal{Y}\_2 + \dots + \omega\_{\mathcal{U}\_N} \cdot \omega\_{\mathcal{Y}\_N} \cdot \partial \mathcal{Y}\_N) \tag{5}
$$

where <sup>ω</sup>*Ui* = <sup>∂</sup>*<sup>W</sup>* <sup>∂</sup>*Ui* and <sup>ω</sup>*Yi* <sup>=</sup> <sup>∂</sup>*<sup>U</sup>* ∂*Yi* . According to the approximation suggested by OECD [50], the equity weight ω for a marginal increase in income for a person with income *Yi* can be computed as:

$$
\omega\_{Y\_i} = \left(^{\chi}/\chi\_{w\_\mathcal{B}}\right)^{-\gamma} \tag{6}
$$

By introducing this equity weight in the calculation of EADs, one obtains an alternative measure, named by Kind et al. [40] Equity Weight Expected Annual Damages (EWEAD). EWEADs are obtained as the product of EAD and the equity weight, and they represent the weight assigned to a dollar loss by the affected individual.

A further alternative measure can be obtained by combining the three approaches above, so as to include both considerations of varying marginal disutility of losses, which may be important when damages are a significant share of incomes and these incomes are unfairly distributed, and of risk aversion, relevant when available compensations are insufficient and, again, distribution of income is significantly unequal. The resulting measure, called Equity Weight Certainty Equivalent Annual Damage (EWCEAD) [40], can be calculated by multiplying the EAD by the equity weight and the risk premium multiplier.

To sum up, the four alternative evaluation methodologies can be expressed as:


In the following sections, we implement them in an empirical valuation of flood damages in our case study, we analyze and compare the results obtained and we highlight the implications of alternative methodological choices.

#### **3. Data**

#### *3.1. The Research Context*

This research was developed in connection with the project "Climatic Resilience of Duran" (RESCLIMA DURAN), to which the University of Turin contributed with a study on the economic valuation of damages complementing the hydrological, geotechnical and community perception analyses developed by local experts (e.g., Tauzer et al. [51]) and by several other European and North American universities and research institutes (a project description is available at: https://www.researchgate.net/project/CLIMATE-RESILIENCE-FOR-CITIES-IN-ECUADOR-Caseof-Duran-RESCLIMA). The Duràn Canton, our study area, is part of the Guayas province in Ecuador, in the estuarine region of the Guayas River (Figure 1). The total area is 331.22 km2, of which 58.14 km2 of urban area and 273.08 km2 of rural area. 97.91 percent of the about 272,000 inhabitants are urbanized. It represents a growing municipality within the largest urban center in Ecuador, Guayaquil, characterized by demographic and socio-economic dynamics—in terms of urbanization trends, segregation between modernized sectors and marginal areas, insecurity, high inequality [52]—typical of large cities in tropical areas.

**Figure 1.** Duràn Canton, Ecuador. (**a**) Map of Duràn urban area; (**b**) Map of Ecuador.

The Canton is composed of 531 census sectors, but the latest Ecuador census (Encuesta Naciònal de Ingresos y Gastos de los Hogares Urbanos y Rurales; Instituto National de Estadistica y Censos (INEC) 2011 [53]) covers only 18 of them. In these sectors, between 10 and 13 families per sector were surveyed, for a total of 213 household observations, which constitute our sample. The survey contains data on population, education level, persons employed, monthly income, monthly expenditure on food and house typology. Houses are classified into four main typologies: villas, independent houses (smaller than villas), apartments in buildings, and houses made of wood or canes. Considering the predominant construction material, houses are further divided in concrete houses, brick-only houses, wooden houses, and cane houses (Table 1).

The average households' annual income is around USD 8000. The sampled houses measure, on average, 68 m2 and are mostly built with concrete (81 percent), although 16 percent of the houses is still made of wood or canes. Out of the 213 household observations, 153 are house owners (72 percent) and the remaining 60 (28 percent) are tenants.

Latitude and the combination of the cold Humboldt current with the hot currents in Gulf of Panama and the El Niño Southern Oscillation (ENSO) phenomenon give Ecuador, with the exception of the Andean regions, a tropical climate, with heavy precipitations between January and May leading to frequent overflows of the Guayas river and the region's inner waterways. Coastal Ecuador is one of the highest hydraulic risk locations in Latin America, and cities along the mouth of the Guayas river rank among the most vulnerable areas to flooding worldwide [54]. The urban area of Duràn Canton is at an altitude varying between 0 and 88 meters above sea level. Unstructured urbanization has pushed the poor into the risk prone lowest-lying areas [51,55].


**Table 1.** Descriptive Statistics (Source: our elaboration on [53]).

#### *3.2. Return Period, Stage-Damage Function and Flood Inundation Map*

According to hydrological models developed by the local government [56], the largest part of the Duràn Canton territory experiences extremely frequent flooding, with estimated return periods of five years (blue area in Figure 2). The most urbanized census sectors are mainly subject to return periods of up to 25 years. Arnell et al. [57] report that the frequency of river flooding in the period 1961–1990 will likely double by 2050 in Central and Eastern Europe, Central America, Brazil and some parts of Western and Central Africa. According to data reported in the EM-DAT database, the average annual number of flood events worldwide has increased from under 30 between 1971–1980 to almost 50 between 1981–1990 to over 140 between 2011 and 2015.

The stage-damage (or depth-damage) function, as mentioned above, is a function that connects damages to the depth of flood water. The database of the Joint Research Center of the European Commission (JRC) created by Huizinga et al. [58] contains damage factors of the function for all Latin American countries. The maximum damage value is estimated for Ecuador in USD 436 per square meter. This value—the highest in Latin America—represents the sum of structural and house or other building content damages, with structural damages estimated at USD 291/sqm and content damages at USD 145/sqm. We have adjusted damage values, as suggested by the JRC guidelines [58], considering rural versus urban context and the predominant material of buildings. The stage-damage function for Latin America is reported in Figure 3.

The values of flood depth in Duràn Canton for a return period of five years, described by the color gradient in Figure 4, were obtained from maps developed by Tapia [59].

**Figure 2.** Return period map for Duràn Canton. (Source: our re-elaboration on [56]).

**Figure 3.** Stage-damage function for Latin America (Source: our adaptation on [58]).

**Figure 4.** Flood inundation map for five years return period (Source: Our elaboration on [43]).

Total damages were derived from the stage-damage function and the inundation maps, for each censual sector and for each return period. Total damages were calculated dividing the house dimensions (square meters) by the return period of floods. The result is the Expected Annual Damage.

#### **4. Empirical Equity Weights and Risk Premium Multipliers**

In order to compute the risk premium multipliers of Equation (2) and the equity weights of Equation (6), we need empirical values for the elasticity of marginal utility (γ) and for the standard vulnerability (*Z*). We compute the equity weights, starting from the annual income per census sector, considering also risk aversion and income distribution. The elasticity of marginal utility, which must be γ > 0 and γ 1, varies across countries and with the level of development. An estimated value for Ecuador is not available in the literature. Existing empirical estimates include Evans [60], who provides an average value of 1.4 in 20 OECD countries; Kula [61], who estimates a value of 1.64 for India; and Lopez [62], who computes the elasticity of marginal utility for nine Latin American countries with values between 1.1 and 1.9, as shown in Table 2.

**Table 2.** Elasticity of marginal utility in Latin American countries (Source: [62], p. 12).


In order to select a value of γ appropriate for Ecuador, we conduct a sensitivity analysis by varying γ in the range 1.1–1.9, the interval of values estimated for Latin American countries by Lopez [62]. The results of the sensitivity analysis are available on request from the corresponding author. The results, in terms of expected damages, remain almost unchanged as the value of γ increases. Then we assume a value of γ = 1.5, considering that the income distribution and the Gini Index in Ecuador are comparable to the ones reported for other countries in South America (e.g., Bolivia, Nicaragua, Mexico) that show elasticities of marginal utility in the range 1.3–1.5 [62]. The resulting equity weights for each census sector are reported in Table 3.

From the latest Ecuador National Survey of Income and Expenditure of Urban and Rural Homes (2011) [53], we retrieved information also on each household status of house owner or tenant, whose descriptive statistics were reported in Table 1.

An important methodological issue highlighted by our Duràn Canton case study, but of high general significance particularly for natural disasters in developing countries, is that standard vulnerability, computed as share of income eroded by annual flood damages (however computed), *Z* = *Flood damages*/*Yi*, is unable to account for damages higher than the annual income. Indeed, in our empirical analysis we find that, in poor neighborhoods, the case of households hit by flood damages to their properties (houses or their contents) higher than the family's annual income is all but infrequent. This implies a term *Z* > 1 and hence a negative risk multiplier: in this way, standard analytical tools truncate the accounting of fractional losses suffered by the poorest.

In order to overcome this limitation, we substitute the share of income lost due to the flood with the fractional value of flood damages over total wealth (*TW*), *Z* = *Flood damages*/*TWi*. If the house is owned, the total wealth includes both income and the damageable value of the house, and potential flood damages are relative both to the structure and the contents. If the house is not owned, potential flood damages can only reach the maximum damage value for the contents, and total wealth is given by the sum of income and the damageable part of the contents. As a proxy of total wealth, therefore, we use the sum of annual income and the maximum value of potential flood damage obtained from the stage-damage function. In the case of households owning their house, the maximum value includes both structural and contents damage (USD 436/sqm); tenant households can only suffer contents damage (the maximum value of which is estimated in USD 145/sqm).


**Table 3.** Empirical equity weights.

The substitution of income lost to flood damages with the share of total wealth lost is an innovation with respect to standard approaches, which allows us to have a value of vulnerability *Z* always between 0 and 1, obtaining valid values for the risk multiplier also for the poorest population quantiles.

The average risk premium multipliers present a slightly rising trend as the return time increases (Figure 5) due to more intense flooding and greater damages to buildings. However, given the peculiarities of our case study, the variability of average risk premium is limited. Figure 5 also shows the census sectors not impacted at low return times (sectors 39002 and 09003), in which the average risk premium is zero.

**Figure 5.** Average risk premium multipliers for census sectors presenting damages. Return periods between 5 and 100 years and γ = 1.5.

#### **5. Results**

To summarize, our empirical analysis combines information on (i) income and house owner or tenant status for the 213 household observations in the Duràn Canton covered by the INEC 2011 census; (ii) damage factors from Arnell and Lloyd-Hughes [57]'s Latin America stage-damage function; and (iii) values of flood depth in Duràn Canton for a return period of five years, from the inundation maps [59]. We compare the resulting evaluation of flood damages obtained with the four alternative methodologies discussed in Section 2, for return periods of 10, 25, 50 and 100 years and under the assumption of a constant elasticity of marginal utility of income of 1.2.

Figures 6–9 display the damage profiles for Expected Annual Damages, Certainty Equivalent Annual Damages, Equity Weights Expected Annual Damages and Equity Weights Certainty Expected Annual Damages, respectively.

**Figure 6.** Damage profile evaluated with Expected Annual Damage (EAD), by census sector and return period.

**Figure 7.** Damage profile evaluated with Certainty Equivalent Annual Damage (CEAD), by return period.

Figures 6–9 show a rapid reduction in the estimated damages as return times lengthen, regardless of the calculation method used. This happens because, in the specific context of the Duràn Canton, flood events are already particularly severe with low return times and they decrease with longer times. In particular, if we look at the case of EAD, which is the ratio between total damages and the probability of occurrence (Figure 6), it becomes clear that if damages do not increase as the return time

increases, the ratio of these two measures will tend to decrease. This result is definitely site-specific and it depends on both the orographic characteristics of the case study and the simulated inundation maps. We also observe that some census sectors are not affected by inundations for return periods of 5 and 10 years but they are with longer periods (i.e., 39002 and 09003).

**Figure 8.** Damage profile evaluated with Equity Weights Expected Annual Damage (EWEAD), by return period.

**Figure 9.** Damage profile evaluated with Equity Weights Certainty Equivalent Annual Damage (EWCEAD), by return period.

Finally, we can notice two main differences among the methods used to compute expected damages. When we take into account income distribution and risk premium, the ranking of sectors by intensity of damage is significantly altered by the choice of evaluation methodology. Moreover, the shape of the curves tends to be more complex when only risk premium multipliers are considered (CEAD in Figure 7) because risk premium multipliers are more heterogeneous among return times and they tend to be more clearly traced when we introduce the distribution of income through equity.

In order to allow an explicit comparison of damage evaluations conducted with the four alternative methodologies, in Table 4 we report the results for all sectors for a return period of five years. Out of the 18 sectors of Duràn Canton, eight are inundated with a return period of five years. The other sectors are never inundated or are inundated for longer return periods: a return period of five years maximizes the area interested by floods (Figure 2).




The area suffering the highest damages is Sector 4002, with total EAD of USD 13,901 and average EAD of USD 1263. Sector 4002 is not the most frequently and severely inundated sector, but it is the sector, along with 6011, with the highest average annual per household income, larger houses and where a bigger share of families are house owners. The predominant construction material is concrete, which makes for houses of higher value with respect to brick-only, wooden or cane constructions more frequent in lower income sectors. Due to the very high value of Expected Annual Damages, Sector 4002 ranks as the most damaged sector also under the CEAD methodology, even though it does not have the highest risk premium multiplier.

However, when equity weights are considered, Sector 4002 is no longer the most impacted sector. Sectors 28008 and 18002, areas with high equity weights and risk premium multipliers, which rank second and fourth respectively under the Expected Annual Damage framework, become the first and third most severely affected areas if equity weights and risk premium multipliers are accounted for in the evaluation of damages (EWEAD and EWCEAD). Conversely, Sector 6011 (the sector with the highest average income per household), which would be considered the second most damaged area under a standard EAD approach, slides down to fourth position in the ranking if damages are evaluated with equity weights. The adoption of methodologies that incorporate information on income distribution does alter significantly the outcome of evaluations and the ranking of target areas for compensation and reconstruction.

In Table 4, we report also the median value for each of the alternative methodologies used to compute expected damages. This measure of central tendency helps us to identify the census sectors presenting low-income households suffering severe damages and, in general, more unequal income distributions. This is the case for Sector 4002, which presents the highest average EAD but is among the sectors with the lowest median EAD; in this sector, the presence of few households with very low annual income exerts a strong effect on the mean which is instead mitigated by the median.

#### **6. Conclusions**

The EAD framework represents the procedure to evaluate damages from natural disasters in a typical CBA. Indeed, standard CBA is a satisfying procedure when adequate schemes are in place for the compensation of damages, income distribution is fair and damages are moderate. However, this is not the case in many instances—particularly in urban areas with low average income and marked inequality. By testing EAD and three alternative evaluation methodologies on data from a particularly significant case study—a coastal tropical urban area among the most vulnerable to flooding worldwide—we provide evidence of general value and a framework replicable in any other relevant context. Our results show that alternative measures of monetary damages from natural disasters, more coherent with economic theory of individual preferences and a social welfare perspective, can substantially modify both compensations and the ranking of priority areas of intervention. Our empirical implementation of the theoretical framework proposed in Adler [39] and Kind et al. [40] shows that the observation of income distribution, specifically via its reflection on marginal utility of income and on risk aversion, may provide a different view from the commonly adopted approach and it allows decision makers to pursue mitigation, adaptation and compensation policies more closely, reflecting a social welfare objective.

Obviously, this study also leaves room for further improvements. We have used a general stage-damage function fitted to Latin American countries, whereas more sophisticated, ad hoc studies could develop specific stage-damage functions fitted to the specific evaluation area—Ecuador or Duràn Canton data, in this case. We have used the latest available census, published in 2011 [53]; the study could be validated and updated by using the new census data which will become available in 2021-22. The sensitivity analysis could be enriched: particularly (i) a specific value of the γ parameter for the area of interest could be calculated from original data; and (ii) the analysis could be repeated with different utility functions. Further studies, replicating the analysis in other contexts and perhaps refined along these lines, would contribute to strengthening the case for revisiting the way CBA is performed

in the presence of high-income inequality. We hope this first empirical investigation will spur further research interest on alternative approaches for the monetary valuation of the impacts of floods and other natural disasters on people's livelihoods.

**Author Contributions:** Individual contributions to the research are as follows: Conceptualization, V.F.; Data curation, F.S.; Methodology, V.F. and S.D.; Supervision, S.D. and A.P.; Writing—original draft, F.S.; Writing—review & editing, V.F., S.D. and A.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was developed in connection with the project RESCLIMA, funded by the Municipality of Duràn and the Escuela Superior Politecnica del Litoral (ESPOL) and executed by the Pacific International Center for Disaster Risk Reduction (CIP-DRR). However, it was self-financed through a co-funding of the University of Turin and the Polytechnic of Turin.

**Acknowledgments:** The authors would like to thank Prof. Mercy Borbor-Cordova, coordinator of the project RESCLIMA DURÀN, colleagues from the ESPOL (Escuela Superior Politécnica del Litoral, Guayaquil, Ecuador) and Angel Valdiviezo from the General Direction of Risk Management of Duràn Canton for sharing local experience, information and data necessary to implement the case study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


62. Lopez, H. *The Social Discount Rate: Estimates for Nine Latin American Countries*; Policy Research working paper WPS 4639; World Bank: Washington, DC, USA, 2008. Available online: http://documents.worldbank.org/ curated/en/135541468266716605/The-social-discount-rate-estimates-for-nine-Latin-American-countries (accessed on 2 September 2020).

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Community Perception and Communication of Volcanic Risk from the Cotopaxi Volcano in Latacunga, Ecuador**

**Juan Camilo Gomez-Zapata 1,2,\* , Cristhian Parrado <sup>3</sup> , Theresa Frimberger 4, Fernando Barragán-Ochoa 5,6, Fabio Brill 7,8 , Kerstin Büche 9, Michael Krautblatter 4, Michael Langbein <sup>10</sup> , Massimiliano Pittore 1,11 , Hugo Rosero-Velásquez 12, Elisabeth Schoepfer 10, Harald Spahn <sup>13</sup> and Camilo Zapata-Tapia 14,15**

	- <sup>9</sup> Geomer GmbH, 69126 Heidelberg, Germany; kerstin.bueche@geomer.de
	- <sup>10</sup> German Remote Sensing Data Center (DFD), German Aerospace Center (DLR), 82234 Oberpfaffenhofen, Germany; Michael.Langbein@dlr.de (M.L.); Elisabeth.Schoepfer@dlr.de (E.S.)
	- <sup>11</sup> EURAC Research, 39100 Bolzano, Italy; massimiliano.pittore@eurac.edu
	- <sup>12</sup> Engineering Risk Analysis Group, Technical University of Munich, 80333 Munich, Germany; hugo.rosero@tum.de
	- <sup>13</sup> Independent Consultant, 26689 Apen, Germany; harald.spahn@web.de
	- <sup>14</sup> College of Social Sciences and Humanities, Universidad San Francisco de Quito, Quito 170901, Ecuador; camilozapatatapia@gmail.com

**Abstract:** The inhabitants of Latacunga living in the surrounding of the Cotopaxi volcano (Ecuador) are exposed to several hazards and related disasters. After the last 2015 volcanic eruption, it became evident once again how important it is for the exposed population to understand their own social, physical, and systemic vulnerability. Effective risk communication is essential before the occurrence of a volcanic crisis. This study integrates quantitative risk and semi-quantitative social risk perceptions, aiming for risk-informed communities. We present the use of the RIESGOS demonstrator for interactive exploration and visualisation of risk scenarios. The development of this demonstrator through an iterative process with the local experts and potential end-users increases both the quality of the technical tool as well as its practical applicability. Moreover, the community risk perception in a focused area was investigated through online and field surveys. Geo-located interviews are used to map the social perception of volcanic risk factors. Scenario-based outcomes from quantitative risk assessment obtained by the RIESGOS demonstrator are compared with the semi-quantitative risk perceptions. We have found that further efforts are required to provide the exposed communities with a better understanding of the concepts of hazard scenario and intensity.

**Keywords:** risk communication; volcanic hazards; social risk perception; resilience; demonstrator; scenario; multi-risk analysis

#### **1. Introduction**

An active volcanic environment is prone to produce cascading and compound natural hazards. Cascading hazards comprise a primary hazard triggering a secondary one [1],

**Citation:** Gomez-Zapata, J.C.; Parrado, C.; Frimberger, T.; Barragán-Ochoa, F.; Brill, F.; Büche, K.; Krautblatter, M.; Langbein, M.; Pittore, M.; Rosero-Velásquez, H.; et al. Community Perception and Communication of Volcanic Risk from the Cotopaxi Volcano in Latacunga, Ecuador. *Sustainability* **2021**, *13*, 1714. https://doi.org/10.3390/su13041714

Academic Editor: Maurizio Tiepolo Received: 31 December 2020 Accepted: 1 February 2021 Published: 5 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

whilst compound hazards refer to events (not necessary interdependent) events whose spatiotemporal footprints overlap (i.e., they occur almost simultaneously and affect the same or neighbouring locations) [2]. For instance, increasing volcanic activity can occur in company with seismic activity and continuous gas emissions and lightning and ultimately trigger lava flow, pyroclastic density currents, tephra (including volcanic ash and ballistics), debris avalanches (sector collapse), tsunami (for submarine volcanoes or at the seaside), and lahars [3]. Syneruptive lahars (also called primary lahars) can happen due to glacier melting during a volcanic eruption, whilst secondary lahars are commonly triggered by heavy rainfalls [4,5]. The forecast of cascading and/or compound volcanic hazards is very diverse and with time dependencies. Models are heavily tailored towards the specific volcanic system [6]. To explore the possible consequences before the actual occurrence of the events, risk scenarios are instrumental for risk communication practises. A risk scenario, as stated in [7], is considered as a situation picture in which a hazardous event with a certain probability would occur and cause some damage. The appropriate communication of risk scenarios can ultimately contribute to territory planning, response planning, design of evacuation routes, and enhancing overall preparedness.

Consequences of volcanic events can be severe, especially when the affected community is not well prepared. One of the most widely known examples of physical damage on assets and human losses due to a lack of effective risk communication occurred during the 1985 eruption of Nevado del Ruiz volcano in Colombia, during which 25,000 people died due to primary lahars [8–10]. However, volcanoes do not only affect the communities in their proximities, but have also generated systemic infrastructure failures and cascading effects (cascading effects can be defined as the disruptions consequent upon the preceding event that can have an acting large-scale across sectoral boundaries [11]) on a large-scale. A clear example of this type of effect occurred during the eruption of the Eyjafjallajökull volcano in Iceland. During two months in 2010, about 100,000 flights between Europe and North America were cancelled due to the sustained ash emission, causing more than \$1.7 billion losses in lost revenues for airlines [12]. A further example of volcanic multi-hazard risk is the 2018 eruption of the Anak Krakatau volcano in Indonesia, which induced its own flank collapse, triggering a tsunami that resulted in the death of 430 people mostly in the western area of Java Island [13]. A tsunami threat from the Krakatau volcano was not unknown, since a similar historical event happened in 1883. However, it was not taken up in a broader discussion on how to deal with such a risk scenario [14].

Monitoring of volcanic activity has been significantly improved in recent years through denser and more widespread networks [15,16]. Moreover, there have been increasing research activities on the interaction between volcanic hazards (e.g., [17,18]). However, the impacts caused by volcanic hazards are rarely assessed in a comprehensive manner due to the lack of worldwide unified exposure models [19] and the scarce damage data collection on the exposed assets needed to constrain vulnerability models [6]. These difficulties are even more pronounced in a multi-risk context, where there is still a gap in the investigation of the interactions at the vulnerability level [20]. Hence, only a few examples of quantitative damage assessment have been reported in the scientific literature (e.g., [21,22]). Furthermore, there is a lack of tools for simulating representative volcanic scenarios in order to analyse the extent and spatial distribution of the expected consequences, needed for decision making and planning. Therefore, scenario-based approaches for a volcanic multi-hazard risk environment are not always available or might not be effectively communicated to the exposed communities before the occurrence of a volcanic crisis [23]. On the one hand, setting up these methods in a consistent scenario-based approach is a challenging task on its own. On the other hand, effectively communicating the potential direct damages and losses and the associated likely disruptions of critical infrastructure is also a daunting task, which in turn depends on the availability of scenario-based risk outcomes.

Rural communities of economically developing countries are particularly prone to encounter more difficulties throughout every single step of the multi-hazard risk chain (e.g., [24,25]). The social vulnerability perception of rural inhabitants might not be always taken into consideration by the local planners, partially due to their remoteness, i.e., typical large distances from the main urban centres [26], or socio-economic factors such as their alphabetization level [27], poor access to information systems, or even the basic lack of knowledge of what potential hazards may impact their communities [28]. These characteristics are common in areas exposed to volcanic hazards. In 2015, roughly 415 million people, most of them located in rural areas, lived within a 100 km radius from the 220 active volcanoes listed in the NOAA Significant Volcanic Eruption Database [29]. Hence, rural communities worldwide are more prone to suffer damaging effects from volcanic eruptions [30]. These consequences are not only expected to impact individual components such as buildings [31] and agricultural fields [32], but also critical infrastructure (e.g., power networks, roads, and water supply systems) for which the evaluation of systemic vulnerability is also required (e.g., [33,34]).

Cascading effects may further drastically change the health quality, as well as economic and social activities of the exposed communities [35]. For example, the continuous emissions of volcanic ashes can interrupt agro-industrial activities, which are the most typical source of income of rural communities [36]. These communities may also experience low serviceability of lifeline networks [37] and/or suffer from physical isolation from neighbouring communities, e.g., due to damaged bridges. Only in a few cases, the cascading effects due to volcanic eruptions have been analysed in a systematic manner [38]. Therefore, there is an urgent need to effectively communicate the scientific results of a volcanic risk assessment while simultaneously addressing the social perception and understandings, by the exposed communities, of different risk factors [39]. As stated in [40], clear risk communication in all the components of a multi-risk chain (i.e., hazards, exposure, and physical and systemic vulnerabilities) with the directly exposed communities, local decision-makers, and planners is fundamental to construct more resilient communities.

Although community participation is considered an essential component of effective resilience planning to natural hazard-risks [41,42], only in recent years some studies have integrated scientific approaches with the active participation of the community, local planners, decision-makers, and actors of the civil society (e.g., [43–47]). The specific community perceptions of vulnerability and risk related to volcanic hazards have been investigated in former works (e.g., [48–51]) through "top-down" approaches. In [52], it has been suggested that scientists should have a transversal role and a stronger presence in the communication of volcanic hazards and risks from "bottom-up" approaches. To the best of the authors' knowledge, these practices have been documented in a few works for rural communities (i.e., [53,54]). Hence, we realise that there is still significant work to be carried out to strengthen the risk-informed communities exposed to volcanic hazards. With this background, we present throughout this work an integrative framework between scientific approaches that study the possible damaging effects from volcanic scenarios with the local knowledge and social risk perceptions. The study area of Latacunga, capital of the Cotopaxi province in Ecuador, with mainly rurally composed communities, and exposed to the Cotopaxi volcano has been investigated in order to enhance a risk-informed community and awareness and contribute to increasing their resilience.

#### **2. Framework and Objectives**

Volcanic eruptions pose an enormous risk to Ecuador, because most of the exposed human settlements in the central and northern highlands are situated less than 25 km from an active volcano. Cities previously affected by volcanic eruptions include Quito, Latacunga, Salcedo, Cayambe, Ibarra-Otavalo, Ambato, Riobamba, and Baños [55]. Lahars have been among the deadliest volcanic hazards, but the emission of volcanic ash has been more frequent in the Ecuadorian Andes [56]. Ash falls do not only have direct consequences on the inhabitants' health and on the exposed infrastructure, but also on agriculture and animal husbandry, which is particularly important for the rural communities in Ecuador. Ash falls have hit the rural communities settled in the vicinity of the most active Ecuadorian volcanoes (i.e., Tungurahua, Reventador, Sangay, and Cotopaxi). Moreover, poverty,

marginality, and high inequality of the exposed communities coexist with their physical and systemic vulnerabilities [57].

#### *2.1. Description of the Study Area*

The Cotopaxi volcano is an active stratovolcano (5897 m.a.s.l) located in the Cordillera Real of the Ecuadorian Andes (Figure 1) and is covered by an extensive, but diminishing glacier cap. Cotopaxi is one of the most dangerous volcanoes worldwide [58] with average recurrence intervals for eruptions between 117–147 years [59]. It can produce syneruptive lahars triggered by explosive eruptions, which can travel hundreds of kilometres [60]. Three drainage systems originate on Cotopaxi (Figure 2), which have all been inundated by lahars in prehistoric times [61]. However, only the northern and southern drainage systems are densely populated: The largest urban agglomeration encountered by the northern system is "El Valle de Los Chillos" (with about 400,000 inhabitants) in the vicinity of southern Quito, whilst the southern drainage system encounters the Latacunga canton (with about 300,000 inhabitants). The last major eruption of the Cotopaxi volcano in the historical records occurred in 1877. It induced syneruptive lahars that severely affected the proximal rural communities [62], with more than 1000 deaths registered, and caused a severe economic crisis [63]. If a similar scenario occurred nowadays, the social and economic consequences would be far more catastrophic due to the high population density and the central importance of Latacunga and the Cotopaxi region for the economic development of the country [58].

**Figure 1.** Location of the main volcanic systems in the Ecuadorian Andes highlighting the location of the Cotopaxi volcano and Latacunga. Modified after [64].

Latacunga is the largest city of the Latacunga canton (second Ecuadorian administrative division) and it is the capital of the Cotopaxi province (first division). It is located at a 14 km distance from the Cotopaxi volcano. For the year 2020, and based on the population projections of the National Institute of Statistics and Censuses [65], the city has an inferred population of approximately 205,600 inhabitants, with a major rural composition (59.8%). The last peak of volcanic activity of the Cotopaxi volcano occurred in mid-April 2015 and lead to a crisis in risk management in Latacunga and neighbouring municipalities [66]. Firstly, an increase in the seismic activity of the volcano was accompanied by the emission

of sulphur dioxide and ash fall for some weeks [56]. Subsequently, authorities and local press communicated to the inhabitants of the communities in the vicinity of the Cotopaxi volcano that it was necessary to evacuate their homes promptly due to the imminent occurrence of lahars [66]. This generated social chaos due to the ignorance of the evacuation routes, the uncontrolled behaviour of the citizens (due to generalised fear of looting), as well as a very low level of trust in government representatives [67]. Eventually, the 2015 activity never surpassed a magnitude VEI 2 and no large syneruptive lahar flows occurred [68]. The lesson learned from this experience was the need for adequate evacuation protocols and local authorities with an understanding of the complexity of the risk in the area. Moreover, it was realized how important it is for citizens to understand their own social, physical, and systemic vulnerability [68].

**Figure 2.** (**a**) Location of the Cotopaxi volcano and the main drainages and populated centres. (**b**) Estimated lahar footprints in the southern drainage system from three scenarios as function of the VEI (Volcanic Explosivity Index). (**c**) Brief description of the eruption scenarios expected at Cotopaxi in terms of the VEI. Modified after [64,69].

Latacunga is settled on ancient and recent geological materials formed by volcanic material. Some of the most representative and better-exposed stratigraphic formations of ancient ashes and lahar deposits originated from the previous volcanic activity of the Cotopaxi volcano were visited (Figure 3) with the guidance of experts from the Geophysical Institute of the National Polytechnic School (IG-EPN (Instituto Geofísico de la Escuela Politécnica Nacional (Quito, Ecuador))) and the Decentralized Autonomous Government of the Province of Cotopaxi (GADPC (GADPC, Gobierno Autónomo Descentralizado Provincial de Cotopaxi, Latacunga, Ecuador)). Some of these deposits are from pre-historical times, whilst the shallower ones date from the 1877 event that destroyed Latacunga [70]. Official maps of the Geological and Energy Research Institute (IIGE) and IG-EPN [71] were used during the field reconnaissance. This field trip was relevant to visualize the geological characteristics of the study area, as well as to strengthen the cooperation and idea exchanges with the local experts.

**Figure 3.** (**Left**): Channel of the Cutuchi River in the city centre of Latacunga. An old textile factory is visible, which has been buried up to the forth story by the 1877 lahar. (**Right**): Thick sequence of lahar deposits, scoria flow deposits, and tephra beds exposed in a quarry along the Rio Saquimala close to Mulalo. (Photos: Theresa Frimberger, 2018).

> Latacunga is not only exposed to the natural hazards imposed by the Cotopaxi volcano, but also to other geodynamic (e.g., landslides and earthquakes) and hydro-climatologic hazards (e.g., frosts and droughts). As reported in [72], there has been an intensification in the variability of precipitations, droughts, and frosts in Latacunga. This has been evidenced in the period between the years 1981–2014, during which the average air temperature has increased about 0.8 ◦C. These ongoing phenomena related to climate change have generated negative consequences mainly in the rural area and in agriculture areas [72].

#### *2.2. Objectives*

The understanding of disaster risk based on the independent investigation of their dimensions, hazards, exposure, and vulnerability, guided by a multi-hazard risk approach with risk-informed decision-makers, is the key advice of the Sendai Framework for Disaster Risk reduction (2015–2030) [73]. Having in mind the aforementioned limitations on volcanic risk assessment as well as the lack of exploration tools for risk communication, two initiatives, namely the programme "Sustainable Intermediate Cities—CIS" and the research project "Multi-Risk Analysis and Information System Components for the Andes Region—RIESGOS" have been working in Latacunga, Ecuador, with the aim of increasing awareness and preparedness and enhancing the coping capacities of the communities exposed to the Cotopaxi volcano. The particular objectives of this integrative study are:


#### **3. Materials and Methods**

An integrative framework between scientific approaches and risk communication practices with the exposed society has been set up in Latacunga (Ecuador) by two different initiatives, (1) the CIS (Sustainable Intermediate Cities) programme and (2) the RIESGOS project (Multi-risk analysis and information system components for the Andes region).

#### *3.1. The CIS Programme: The Creation of a Local Laboratory to Evaluate the Social Perception of Risk and Resilience*

"The Latacunga Laboratory: Risk management, resilience, and adaptation to climate change (Laboratorio Urbano de Latacunga: Gestión de riesgos, resiliencia y adaptación al cambio climático)" has been created within the CIS programme, as part of the joint initiatives of GIZ ("Deutsche Gesellschaft für Internationale Zusammenarbeit") and Grupo FAR (Ecuadorian NGO (https://grupofaro.org/ (accessed on 1 January 2021)). The creation of so-called resilience observatories for exposed communities to natural hazards is a relatively new trend [45,74]. Similarly, the Latacunga Laboratory seeks to contribute to the risk management to natural hazards that are likely to occur in the territory, while aiming to contribute in the long-term to the development of the city embracing its urban–rural ties. With that goal, initial contributions related to social risk perceptions have been documented in [75] as a joint effort between the Latacunga Laboratory, the local government, academic institutions, and local actors.

3.1.1. Comparative Analysis of the Social Risk Perception Factors to Natural Hazards and the Spatial Distribution of Volcanic-Related Risk Factors

We conducted a survey by means of a custom-designed questionnaire, a fundamental tool for acquiring information on public knowledge of the community [76]. It is composed of a series of multiple-choice questions in Spanish. The survey was carried out in the field and online to collect data about the individual knowledge, attitudes, and risk perceptions of the inhabitants of Latacunga. The online survey was promoted on social media and was available on the official website of the CIS Latacunga Laboratory (https://latacungaresiliente.com/ (accessed on 1 January 2021)) for a month. In the meantime, the field survey was carried out only in the urban agglomeration of Latacunga. The collected data is used for two main objectives: (1) as input to perform the semiquantitative method proposed in [77] that ranks the social perception of volcanic risk factors (i.e., hazard recurrence, exposure, vulnerability, and resilience) among other natural hazards likely to occur in the study area (i.e., earthquakes, drought, frost, floods, and landslides), and (2) to map the spatial distribution of volcano-related risk perception into comprehensive categories (i.e., easily understandable by the exposed communities).

A design of the field surveying site was carried out. According to the last official census available [65] and population projections by the survey elaboration date (September 2019), 50,442 inhabitants over the age of 18 years were considered as qualified informants. In order to get a statistically representative sample, a confidence level of 95% and a margin of error of 5% were selected. On this basis, we estimated that a sample for the field surveys not smaller than 380 inhabitants had to be selected. Considering 10% additional surveys, a final sample size of 420 people was chosen. The population density (Figure 4a) was used to constrain the spatial distribution of the field surveys within the urban blocks with a residential occupancy (Figure 4b). The surveys were carried out by 55 students of the ISTC (Instituto Superior Tecnológico Cotopaxi) in September 2019.

**Figure 4.** (**a**) Population qualified for the survey to evaluate the social risk perception in the urban centre of Latacunga. (**b**) Sites to survey within the residential buildings. Modified after [75].

It is worth mentioning that, on the one hand, some drawbacks have been found when the community perception of exposure, vulnerability, and resilience are independently addressed for large-scale studies [78,79]. On the other hand, there have been also reported benefits of this separation for mapping the social risk perception to natural hazards (e.g., [80,81]) when bottom-up approaches are carried out. Therefore, we have decided to independently investigate the social perceptions towards these components through separated questions. The Likert scale is used in this context to obtain a quantifiable level of perception of each risk factor. An integer numerical score (1, 2, or 3) is assigned to every possible answer. Although the passage from a qualitative perception to an index can be questioned, several recent studies have shown the usefulness of the Likert scale [82–86]. Notably, in [87], it was found to provide a good compromise between the quality of the information collected and the accessibility to respondents, while the bias in responses decreases, and there is consistency across different measurements and research domains of disaster risk reduction.

Subsequently, the average is computed for every question to obtain the perception of every component. These values are inputs to the computation of the risk perception pre-index through the use of Equation (1), where *P* stands for "perception". An example subset of the questions is presented in Table A1 (Appendix A). The questions and answers were validated by local risk management experts from the Association of Risk Management Professionals of Ecuador (Asociación de Profesionales de Gestión de Riesgos de Ecuador, APGR).

$$P(Risk) = \left(\frac{P(Haxard) \times P(Exposure) \times P(Vulnerability)}{P(Resilience)}\right) \tag{1}$$

The numerator of Equation (1) can have a maximum possible value of 27, whilst the minimum for the resilience term in the denominator is 1. Therefore the maximum risk perception value that this method admits is 27. The values in the range from 1–27 form a "pre-index". To obtain a more comprehensive numerical value, a "reduced index" in the 0–3 range is obtained through the application of Equation (2).

$$Reduced\ index = \log\_3(preindex\ value)\tag{2}$$

The relations between the "pre-index" and the "reduced index" is shown in Figure 5. For mapping purposes an "equal interval" classification for the reduced index scale is introduced with five classes of length 0.6 for finally presenting the spatialized perception of every risk factor in a compressive manner to the community. The calculated results for every answered question at each survey location (Figure 4b) are used to map the spatial distribution of the perception of hazard recurrence, exposure, vulnerability and resilience, and the risk index (computed with Equation (1). Subsequently, they were interpolated through the use of the ordinary kriging geostatistical algorithm [88].

**Figure 5.** Graphical scale and correspondence between the pre-index value and the reduced index.

#### *3.2. The RIESGOS Project: Iterative Simulation Improvement and Enhanced Communication*

The idea of constructing a web-tool, the RIESGOS demonstrator, as a decentralised and intraoperative environment for the exploration of the consequences in Latacunga from different volcanic hazard scenarios was proposed to the local stakeholders who participated in four participative workshops. Two of them were held in Latacunga (7 December 2018; 25 November 2019) in the headquarters of GADPC (Decentralized Autonomous Government of the Cotopaxi Province) and two workshops took place in Quito on 11 December 2018, and on 27 November 2019, respectively. The participants ranged from research partners, representatives of the rural municipalities (parishes) of the Cotopaxi province, public authorities, environment secretaries, actors of the civil society such as local representatives of agriculture associations, and urban and rural leaders. Similarly, as recently presented in [46], the workshops were used as a means to implement a user-centred iterative approach, seeking a continuous redesign of the RIESGOS demonstrator that has been guided by the needs of potential users and practical applicability. This has been ensured by a comprehensive analysis of user requirements (e.g., open-source, user-friendly graphical user interface and transferability).

#### 3.2.1. The RIESGOS Demonstrator Tool for Quantitative Multi-Risk Analysis

The iteratively constructed RIESGOS demonstrator for a multi-risk information system is based on a modular and scalable concept in which the different hazards, the related exposure models, and vulnerability schemas are each represented by one individual web service. These independent and distributed web-services (managed and maintained by individual research institutions) are based on the quantitative methodologies developed within the RIESGOS framework for multi-risk analysis (i.e., [69,89–94]). Therefore, their integration into the RIESGOS demonstrator simulates the multi-risk environment of Latacunga. This modular approach offers the possibility to integrate different web services into already existing system environments.

Currently, the graphical user interface of the demonstrator can be accessed from a web browser only by users with special rights. The main screen of the graphical user interface is divided into three main display areas: the central map window, the configuration wizard for the control of each web service to the left, and the results panel to the

right (e.g., see Figure 6). The code of the graphical user interface (RIESGOS frontend) is openly published on GitHub (https://github.com/riesgos/dlr-riesgos-frontend (accessed on 1 January 2021)). The use of standardized web services such as geospatial web services defined by the Open Geospatial Consortium (OGC) allows users accessing open and flexible multi-risk information and data products. Web-services and exposed data resources can be accessed using a variety of means from a simple command-line tool, over a web browser, to existing graphical user interfaces of public authorities and companies, which are equipped with a map user. OGC web services allow all kinds of geospatial functionality out-of-the-box including data access, data display, styling, and processing. Web services can easily be integrated into existing clients. The providers of web services define their products, display options, and configuration items. More details of this integrative process are reported in [94]. Through the clear separation in competencies between web services and user-interface, modularity and scalability are increased.

**Figure 6.** Example of the graphical representation of loss distribution due to ash fall scenario in the RIESGOS demonstrator (as of December 2020) from a previously selected VEI. Reddish and greenish aggregation areas representing higher and lower values, respectively. On top of these results, the lahar model (with the same VEI) is displayed as input to calculate the cumulative damage over the same geo-cells exposed to both perils.

Precomputed hazard models of ash-falls and lahars are displayed by the RIESGOS demonstrator after the selection of a scenario in terms of the expected for an eruption of the Cotopaxi volcano. Local probabilistic ashfall models for the Cotopaxi volcano generated by the IG-EPN (following the method of [95] with 20-year observation of wind flow directions) are currently integrated as twelve explorative scenarios. They are represented by isolines (Figure A1). The lahar models described in [69] are incorporated, showing the maximum possible values of five physical properties (i.e., flow velocity, flow depth, pressure, erosion, and deposition (see Figure A2).

The exposure model provides the input to calculate the direct losses over residential building portfolios classified in specific building classes for every hazard. An example for lahar-building classes is depicted in Figure A3. These models were constrained through the use of taxonomic characteristics available in the official cadastral dataset of the GADPC (Gobierno Autónomo Descentralizado Provincial de Cotopaxi, Latacunga, Ecuador), such as roof and wall materials, and the proportions of the predominant building materials suggested for Latacunga in [96]. No further details are provided on the manner the building exposure models were constructed, since this is out of the scope of this paper.

The vulnerability analysis of the typical residential buildings is performed using representative building exposure models with their respective fragility functions and suitable economical consequence models. Specifically, this approach is an extension of the Performance-Based Earthquake-Engineering (PBEE) method developed by [97], which has more recently been adapted to other kinds of natural hazards. The fragility model proposed in [98] is used in lahar fragility, whilst the one in [99] is used in ash fall fragility for typical residential buildings that can be encountered in the study area. The demonstrator ultimately obtains the spatial distribution of damage and losses per individual hazard, plus the option of obtaining the cumulative damage and losses due to the action of both hazardous events using the novel method outlined in [90]. Some examples are depicted in Figures 6 and A4. No further technical details are provided, because it is out of the scope of this work. Furthermore, the demonstrator enables the visualization of the areas that might potentially get disconnected from different networks and thus the identification of cascading effects on the economic activity. The method of implementation in the systemic vulnerability analysis applied in this case is similar to the one proposed in [100]. This information can be related with census data for estimating the population that might be affected by a blackout [101]. One example of this process is depicted in Figure A5 for the interruption probabilities of the electrical power network due to the impact of a lahar.

#### **4. Results**

#### *4.1. The Recognition of the Latacunga Local Laboratory by the Local Actors of the Community*

"The Latacunga Laboratory: Risk management, resilience, and adaptation to climate change" has strengthened its presence in the territory through several continuous participative activities that are aligned with the objectives mentioned in Section 3.1. For instance, the Laboratory has been recently working in materialising initiatives that were proposed by local entrepreneurs. One of them is currently working on the recovery of "Relatos de una erupcion" (Tales of an eruption), which works on rescuing the historical memory of what happened in the eruption of the Cotopaxi volcano in 1877. This has been carried out through audio-visual stories that are told by direct descendants who survived this event. This initiative enhances co-responsibility and respect for historical memory. The oral transmission of this information is an important input to generate awareness. Details about these initiatives can be found in the Latacunga Laboratory website (https://latacungaresiliente. com/rescate-de-la-memoria-historica-de-la-erupcion-del-volcan-cotopaxi/ (accessed on 1 January 2021)).

Comparative Analysis of the Social Risk Perception Factors to Natural Hazards and the Spatial Distribution of Volcanic-Related Risk Factors

The method described in Section 3.1.1 was applied to rank the volcanic risk perception for the most densely populated area in Latacunga conurbation. Making use of the 420 processed surveys as input data, the social perception to the recurrence of hazards, exposure, vulnerability, and resilience for six natural hazards likely to occur in Latacunga (i.e., earthquakes, volcanic eruptions, droughts, frosts, landslides, and floods) has been investigated. This is presented in the form of the comparative matrix shown in Table 1, which reports the mean values (for all the surveys) related to the perception of every component, as well as the computed risk index for every considered hazard. The higher the value, the greater the perception of risk. In the case of resilience, the interpretation is the opposite: the higher the value, the higher perception of resilience after a hazardous event.

The greatest concern among the inhabitants of Latacunga is their own perceived vulnerability to volcanic hazards. Remarkably, their resilience after a volcanic event scores the lowest value. This implies the community is aware that they would have great difficulty (or impossibility) to recover from the related damages. It is worth noticing that despite the fact that in the questionnaires there was no distinction made in terms of the type of

volcanic hazards (lahar, ash fall/tephra fall, or ballistics) or in terms of their intensity, the collective imaginary always tended to associate the occurrence of a destructive lahar as "the volcanic hazard". Most likely, the oral transmission of the experiences of the survivors from the 1877 event has permeated the mental construction of their descendants.


**Table 1.** Hazard matrix and perception of risk factors towards natural hazards in the urban area of Latacunga.

The mean results in terms of the percentage for the answered questionnaire that makes up the field and online surveys are depicted in Table A2. Contrary to the field survey, the online surveys score large values in the basic knowledge and reconnaissance of their exposed environment (i.e., evacuation routes, emergency committee, the existence of initiatives for risk reduction). The field surveys express that 64% of the inhabitants consider the volcanic related hazards as events that are likely to happen in the city within their lifetimes. Furthermore, 86% answered that they believe an eventual eruption of the Cotopaxi volcano would cause very serious damaging effects to the city. Likewise, 75% considered they will have very serious impacts directly on their families and themselves. 25% of the population considers that recovery from a serious volcanic event would be impossible, whilst 57% think it would be difficult to overcome. Regarding knowledge, 67% of the population know safe places in the event of a possible disaster, while 61% know evacuation routes. However, only 34% ensure there are emergency plans in their neighbourhood. Half of the respondents do not even know if they live in a volcanic hazard zone. Additionally, 56% of the field-surveyed inhabitants and 69% of the online-respondents consider they would have rapid reaction capacities. Finally, ~42% of the population talks about how to act in case of emergency with their families.

Every answer of the 420 field surveys was spatially distributed onto the survey locations (Figure 4). Their associated numerical values of the Likert scale were interpolated through the use of the ordinary kriging geostatistical algorithm [88]. Subsequently, every numerical value is converted to the equivalent categories presented in Figure 5. The spatially explicit categories represent the social perception of volcanic hazard recurrence, exposure, vulnerability, and resilience in the study area. They are respectively depicted in Figure 7a–d. The former factors are integrated through Equation (1) to generate Figure 7e, which represents the semi-quantitative volcanic risk perception index proposed in [77]. In general terms, the perceptions of hazard recurrence, exposure, and vulnerability are quite similar. However, in the central–easternmost and northernmost zones, there is a high perception of hazard recurrence, a very low perception of resilience, and a moderate perception of vulnerability, whilst in the southernmost part (where the Cutuchi River flows), the four assessed factors show high and very high values that ultimately lead to a generalized "very high" category in the volcanic risk index. This is contrary to what is observed in the central–western and northern parts. Due to the increasing distances from the main drainages, there is a strong anti-correlation between the higher resilience levels (reddish areas) and the other risk factors. Hence, despite the last volcanic crisis in 2015, there is still a generalised very low perception among the inhabitants that they cannot suffer any direct impact or damaging effect after increasing volcanic activity, because they consider the occurrence of lahars (within their lifetimes) to be impossible. Clearly,

the inhabitants of that sector are not aware of the large intensities the Cotopaxi volcano can achieve (e.g., a Plinian activity (VEI > 4) in Figure 2).

**Figure 7.** *Cont*.

**Figure 7.** Spatial representation of the perception of volcanic hazards in the urban centre of Latacunga in terms of (**a**) recurrence; (**b**) exposure level; (**c**) vulnerability; (**d**) resilience; (**e**) risk index calculated using the former components as inputs. Modified after [75].

#### *4.2. The Commnuication of the Scenario-Based Risk Assessment Concept with Local Stakeholders*

During the four RIESGOS participative workshops, the invited stakeholders expressed interest in understanding the impacts of an extreme volcanic eruptions on the exposed elements such as buildings and critical infrastructure. Brainstorming exercises were carried out during the two first workshops. The participants were invited to imagine a future potential eruption with the emission of ash fall and occurrence of lahars. Thereafter, based on their perspectives and local knowledge, it was asked which physical, systemic, and cascading damaging effects they would expect on their built environment, infrastructure systems, and socioeconomic activities.

Some basic concepts of the probabilistic method, as an open-source web-service assesses the vulnerability of the exposed residential buildings (see Section 3.2.1), were presented to the local stakeholders. Due to the iterative approach used in constructing the demonstrator, some of the details that have been presented in this work as methods are actually the initial outputs of the first participative workshops. In this regard, the adaptability of "foreign" lahar vulnerability models (i.e., not developed for Ecuador) in the study area (e.g., [21,31,102]) was initially discussed with the representatives of the scientific local institutions. Due to the absence of locally developed ash fall vulnerability models for the residential buildings in the surroundings of the Cotopaxi volcano, the use of vulnerability models for the southern Colombian Galeras volcano [99] was perceived suitable to be implemented in the risk calculations rather than the fragility functions frequently developed for other areas (e.g., Italy [103]). With this feedback, the web-tool was redesigned. This is an example of how the engagement of local participants can improve both the technical development of quantitative methods (by agreeing on a proper model) as well as the understanding of such methods by the community.

Possible cascading effects that would occur in the case of critical infrastructure failure were debated. For instance, the participants realised that assessing the vulnerability of electric networks to ash falls and lahars is fundamental because of the further consequences on daily social and economic activities. However, the most debated topic was the reliability of the road system that, in the case of failure, may induce physical disruption and affect evacuation and emergency response during a volcanic crisis. Other public infrastructures that would be affected by Cotopaxi's lahars include the Army headquarters "Brigada Patria", Latacunga hospital, and the new penitentiary [58]. The interest in relocating some of the exposed assets was discussed.

"Hands-on" sessions took place during the two last workshops. The participants could experience on their own the use of the RIESGOS demonstrator. They selected different scenarios to visually compare every hazard footprint and intensity (i.e., for ash fall and lahars) as well as their associated risk outcomes on residential buildings and electric power networks. This was done through the selection of individual and successive hazard scenarios addressing cumulative damage. During the "hands-on" session, the participants recognized the potential of the demonstrator as an exploration tool for risk communication.

#### **5. Discussion**

The CIS and RIESGOS projects have independently addressed the domain of risk communication in Latacunga (Ecuador) at different geographical scales. The investigation and mapping of the perception of volcanic risk factors led by the Latacunga Laboratory (created by CIS) was carried out in a focused area (urban area) due to the necessity of having control points (where field-surveys were carried out) for a further geostatistical interpolation process, whilst, in the framework of the RIESGOS project, the construction of the hazard, exposure, and vulnerability approaches for scenario-based multi-risk calculations have been carried out at the canton level. Despite that, the community perceptions of the entire canton and province can be assessed in the future through field surveys for other urban centres (e.g., Pujili, Saquisili, and Salcedo), a meaningful spatially explicit perception of volcanic risk factors could only be mapped for the urban centres. This is because, due to the scattered location of the residential buildings in the rural areas, conventional geostatistical interpolation algorithms would carry significant bias in the results. For the commonly investigated area by RIESGOS and CIS, we can see that the exposed community recognise that they are under a variable level of risk regarding volcanic events depending on their location. These perceptions match the lahar footprints from the scenarios with higher probabilities of occurrence (VEI < 3). However, for larger intensities, (e.g., lahar footprints from a VEI > 4 scenario, see Figure 2b), we observe a mismatch with the spatially explicit community perceptions of volcanic risk factors (Figure 7). For instance, the easternmost areas of the urban centre of Latacunga show low and very low reconnaissance of volcanic risk factors due to their increasing distance respect to the main drainages. The inhabitants of that particular sector have perceived as impossible the occurrence of and suffering from consequences of lahars. The ignorance of the lahar footprints expected from these large intensity scenarios means that the concepts of "safe place" and evacuation routes are not applicable for either. These results should not be interpreted as fixed or permanent, but they rather constitute a temporal reading of the collective mental construction of the inhabitants at the time the surveys were carried out. Nevertheless, considering that the community is placed in ancient lahar deposits, as well as the relatively short time since the last 2015 volcanic crisis, one can realise that from the comparison of the respective outcomes arises the need to prioritize some zones where further divulgation activities should be made in the future regarding the possible scenarios and intensities that the Cotopaxi volcano can actually produce.

The formulated questions comprised in the survey forms are locally revised by experts from the APGR while paying attention to the use of collectively known terminology and the cultural characteristics of the community. In this work, we have implemented a simple numerical expression (Equation (1)) that equally ranks the risks factors of the

different volcanic risk factors. This selection carries epistemic uncertainties. For instance, a customisation weighting schema to each factor, the selection of the median or mode instead of the mean value (herein adopted), together with a broader range in the Likert scale (e.g., 1 to 7 as explored in [104]) could be alternative approaches to be compared or even integrating each other into condition trees as proposed in [105]. The selection of the Likert scale to rank the answers and ultimately map the community perceptions implied an ordinal scale that was further converted into a nominal one based on the "equalscale" (Equation (2)). This decision was made because, since the methods and results are aimed to be divulged, the categories have been found to be comprehensive, easily understandable, and culturally accepted by the community. Although the Likert scale has been extensively and recently used to successfully assess the community perception (e.g., [82–86]), there are several limitations in its adoption. For instance, as stated in [87], this kind of scale, despite maximizing the reliability of answers, also sacrifices the level of detail. However, it should be noted that through the simple possible answers related to the vulnerability perception and the nominal categories, we are only proposing a very simple categorization. More robust approaches that have addressed spatial multi-criteria analysis (as presented in [106]) have shown the impact of addressing diverse socioeconomic variables that we have not addressed in in our approach. A similar situation occurs with the resilience perception, which as discussed in [24], can be decomposed into very heterogeneous variables in economically developed countries.

Therefore, we are not claiming that our results related to the community perception of risk factors are exhaustive, but instead, they should be used as a basis for developing more complex analyses in future stages. For instance, even though we have already observed clear behaviour differences between the responses from online and field surveys, with explicitly designed survey and accounting variables such as work location, alphabetisation level, and economic activity, we could in the future classify the population into different social groups and find similarities and differences in their behaviour within a social environment to carry out more sophisticated methods, as proposed in [107]. Thereby, for each group, we could expect different reactions to a future volcanic crisis and then propose particular resilience practices. However, these kinds of approaches will largely depend on the data availability, which is particularly difficult in the rural tropics [25,57].

As described in recent participative experiences to assess the community perception to natural hazards (e.g., [45,47]) we have also experienced that the workshops carried out were allowed to go beyond a simple exchange of information. They paved the way for a better divulgation of concepts such as triggering and cascading hazards, dynamic vulnerability, cumulative damage, and cascading effects. These understandings in turn facilitated the knowledge flows and feedback acquisition to continuously design the RIESGOS demonstrator guided by increasingly risk-informed decision-makers. With this bottom-up iterative approach in the web-tool design, we are following the suggestions of the Sendai Framework for Disaster Risk Reduction (2015–2030) [73]. The outcomes of the demonstrator are not static hazard maps that are delivered to the exposed population from top-down approaches (e.g., [48–50]), but rather scenario-based online computations that can dynamically change based upon the continuous integration of local datasets and models.

During the "hands-on" sessions, the potential users perceived the RIESGOS demonstrator to intended prompt risk communication processes. For the study area, only hazard models have been typically available, and the few risk outcomes obtained in the past have been reported in tables and not in a spatially explicit manner [58]. Therefore, this work is providing the community with the availability of scenario-based risk models based on the vulnerability of the exposed elements in graphical and user-friendly interphase, which is an added value for the local community. The integrated scenario-based lahar footprints per VEI [69] and the locally developed probabilistic ash falls models [95] are themselves useful outcomes for civil protection and local-planners. They can be used to identify which human settlements and agricultural plantations might be affected or even discuss the relocation of some of the exposed components of critical infrastructure. Although we

have not accounted for the conditional probabilities between triggering and cascading hazards as proposed in [17], we have instead presented fixed risk scenarios. For such a purpose, the demonstrator is served by a novel method that calculates and disaggregates the cumulative damage when there are interactions at the vulnerability level. In the specific volcanic context, although the concept of dynamic vulnerability had been already theoretically sketched in the work of [108], to the best of the authors' knowledge, we have first presented an example case of cumulative damage for risk-informed communities exposed to compound and cascading volcanic hazards. This is an innovative approach that not only contributes to reducing the generalized gap in the interactions at the vulnerability level [35], but also to communicating the results to the local stakeholders. With these contributions, the potential users could identify the most vulnerable areas for further mitigation strategies. It is worth mentioning that, since the RIESGOS demonstrator is currently not an operational tool, but rather shows the scientific and technological capabilities, the economic loss estimations for every exposure geo-cell (where residential buildings are aggregated) should not be used as definitive results. Therefore, due to the underlying uncertainties in these results, there is still the permanent necessity pointed out in [39,52] of having expert local users and scientists who can analyse and effectively communicate this information.

The technology transfer of the activities included in the CIS and RIESGOS programmes is highly relevant. The modular software architecture is particularly relevant for this aspect, for which the databases and methodologies of local Ecuadorian institutions may be ultimately integrated. However, the applicability of the demonstrator in the long-term will depend on how the local authorities will "give life" to the initiative, considering the local legal aspects. For future communication initiatives, due to the intrinsic interoperative sequence of inputs and outputs, the demonstrator can be a pedagogic tool to divulge multi-risk situations as similarly carried out by audio-visual approaches (e.g., [53,109]). Nevertheless, these kinds of local actors should be the first ones to understand the aforementioned concepts of "scenario" and "intensity" within the multi-risk chain, and most importantly, that they can be further contrasted with future and continuous spatially explicit social risk perceptions monitoring initiatives.

#### **6. Conclusions**

We have presented an integrative framework of qualitative community risk perceptions (carried out by the CIS Latacunga Laboratory) and scenario-based quantitative multi-hazard risk assessment (developed by the RIESGOS project). These initiatives have jointly worked on comprehensive volcanic risk communication processes in Latacunga, a city with a mainly rurally composed population, exposed to volcanic hazards from the Cotopaxi volcano.

Online and field surveys were carried out to rank the volcanic risk factors to investigate the individual knowledge and attitudes in Latacunga. Only the geo-located interviews in the field were used to map the community risk perceptions and to calculate a spatially explicit risk perception index through a semi-quantitative approach.

The participative workshops allowed the potentially affected communities to identify how their exposed assets, depending on their physical and systemic vulnerabilities, would be differently affected by several volcanic hazard scenarios. The iteratively customised RIESGOS demonstrator proved to be a useful tool for the communication of quantitative risk scenarios, raising the awareness of potentially affected population for the concept of scenarios and intensity. Its outcomes facilitate discussions among the participants on topics such as relocation of critical infrastructure elements. The demonstrator is not only enhancing the awareness of the communities, but also the user involvement in its development, improving the quality of the software. Although the development of the CIS and RIESGOS methodologies started independently, the respective outcomes of this collaborative work has allowed identifying areas where risk perception and scenariobased risk models are in disagreement. Thus, the need to continue assessing the social

risk perception along with future risk communication efforts in the Cotopaxi region is highlighted.

**Author Contributions:** Conceptualization: J.C.G.-Z., M.P., and C.P.; methodology: M.P., C.P., E.S., F.B.-O., and H.S.; software: M.L.; validation: T.F., C.Z.-T., and C.P.; formal analysis: J.C.G.-Z., F.B.-O., and C.P.; investigation: T.F., F.B., K.B., and M.K.; resources: C.Z.-T.; data curation: J.C.G.-Z., M.L., and H.R.-V.; writing—original draft preparation: J.C.G.-Z., M.L., and C.P.; writing—review and editing: T.F., F.B., and J.C.G.-Z.; visualization: T.F., F.B.-O., and M.L.; supervision: H.S.; project management: E.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research and development project RIESGOS (Grant No. 03G0876) is funded by the German Federal Ministry of Education and Research (BMBF) as part of the funding programme "CLIENT II—International Partnerships for Sustainable Innovations". The development of the Latacunga Laboratory: Risk management, resilience and adaptation to climate change (within the program "Sustainable Intermediate Cities—CIS) is funded by the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) programme of the GIZ Federal Ministry for Economic Cooperation and Development of Germany (BMZ).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the continuous development of the RIESGOS demonstrator and community risk perception carried out by GIZ.

**Acknowledgments:** The authors want to express their gratitude to the 55 students from *Instituto Superior Tecnológico Cotopaxi* (ISTC) who carried out the field surveys, as well as to all the participants of the workshops and the inhabitants in Latacunga who replied the surveys. Thanks to Daniel Straub (TUM), Jörn Lauterjung, Heidi Kreibich, and Fabrice Cotton (GFZ) for their advice during the elaboration of this work. Special thanks to Benjamin Bernard and Sebastian Averdunk (TUM) for the ash fall simulations inputs, as well as to Daniel Andrade and Patricia Mothes (IG-EPN) for the valuable feedback throughout the development of this work. Thanks to Daniela de Gregorio (UNINA), Roberto Torres-Corredor (SGC), Susanna Jenkins (EOS), and Robin Spence (Cambridge A.R) for having kindly provided sets of ash fall fragility functions. Thanks to Karl Heinz Gaudry (GIZ/CIM), Martin Cordovez Dammer, Marta Correa, and Edwin León (IIGE) for the discussions about critical infrastructure and cascading effects during the former volcanic crisis in the study area during the author's visits to Ecuador. Thanks to Hugo Yepes, Pablo Palacios, Jose Marrero, and Juan Carlos Singaucho (IG-EPN) for the feedback about exposure modelling and loss visualization. Thanks to Dorothea Kallenberger (GIZ), Ana Patricia (Grupo FARO), and Cristopher Velasco (president of the APGR) for having supported the research of the community risk perception study in Latacunga. Thanks to Luis Chasi for leading the initiative "Relatos de una erupcion". Thanks to Diego Molina (GADPC) for promoting discussions about the social risk perception of the local actors of the Cotopaxi province during the workshops held in Latacunga. Thanks to all SENESCYT and SNGER for the co-organization of the workshops held in Quito. Special thanks to Nils Brinkmann and Matthias Rüster (GFZ) for having participated in the construction of some of the computer codes implemented in this work.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

#### **Appendix A**


**Table A1.** Example of the procedure for calculating the risk perception pre-index.

**Table A2.** Questionnaire within the survey to assess the social risk perception to volcanic risk in the urban area of Latacunga. The mean values of the entire survey are reported. Adapted after [75].



**Figure A1.** Example of the graphical representation of the spatially distributed ash fall intensities as isolines in the RIESGOS demonstrator (as of December 2020) from a previously selected VEI. The thickness values are displayed. Once a point within the isolines is clicked, the expected load (kPa) value is also shown.

**Figure A2.** Example of the graphical representation of the footprint and intensities of the lahars in the RIESGOS demonstrator (as of December 2020) from a previously selected VEI. On the top-right side of the window, the outputs of the lahar simulation are listed (i.e., lahar flow velocity, flow depth, pressure, erosion, and deposition).

**Figure A3.** Example of the graphical representation of the residential building exposure model in the RIESGOS demonstrator (as of December 2020). It is represented into aggregation areas based on the official rural and urban administrative divisions of Latacunga. There are displayed the quantities of every ash fall risk oriented building class proposed in [99] within a selected area.

**Figure A4.** Example of the graphical representation of damage state distribution due to the combined effect of ash falls and lahar scenarios in the RIESGOS demonstrator (as of December 2020) calculated using the method proposed in [90], [92].

**Figure A5.** Example of the visualization of the expected interruption probabilities in the RIESGOS demonstrator (as of December 2020) of the electrical power network due to the action of a lahar scenario.

#### **References**


#### *Article*

### **Community and Impact Based Early Warning System for Flood Risk Preparedness: The Experience of the Sirba River in Niger**

**Vieri Tarchiani 1,\* , Giovanni Massazza <sup>2</sup> , Maurizio Rosso <sup>3</sup> , Maurizio Tiepolo <sup>2</sup> , Alessandro Pezzoli 2, Mohamed Housseini Ibrahim 4, Gaptia Lawan Katiellou 5, Paolo Tamagnone <sup>3</sup> , Tiziana De Filippis <sup>1</sup> , Leandro Rocchi <sup>1</sup> , Valentina Marchi <sup>1</sup> and Elena Rapisardi <sup>1</sup>**


Received: 7 February 2020; Accepted: 26 February 2020; Published: 28 February 2020

**Abstract:** Floods have recently become a major hazard in West Africa (WA) in terms of both their magnitude and frequency. They affect livelihoods, infrastructure and production systems, hence impacting on Sustainable Development (SD). Early Warning Systems (EWS) for floods that properly address all four EWS components, while also being community and impact-based, do not yet exist in WA. Existing systems address only the main rivers, are conceived in a top-down manner and are hazard-centered. This study on the Sirba river in Niger aims to demonstrate that an operational community and impact-based EWS for floods can be set up by leveraging the existing tools, local stakeholders and knowledge. The main finding of the study is that bridging the gap between top-down and bottom-up approaches is possible by directly connecting the available technical capabilities with the local level through a participatory approach. This allows the beneficiaries to define the rules that will develop the whole system, strengthening their ability to understand the information and take action. Moreover, the integration of hydrological forecasts and observations with the community monitoring and preparedness system provides a lead time suitable for operational decision-making at national and local levels. The study points out the need for the commitment of governments to the transboundary sharing of flood information for EWS and SD.

**Keywords:** early warning; flood risk; hydrology; local communities; Niger river basin; rural development; Sahel

#### **1. Introduction**

As clearly stated by Ban Ki-moon, United Nations Secretary-General, on 1st September 2015: "The Sendai framework has important implications. It shifts the emphasis from disaster management to

disaster risk management." This paradigm shift puts an emphasis on understanding the risks as the underpinning drivers for investing in resilience and preparedness, rather than in response and recovery.

Early Warning Systems (EWS) are a pillar of Disaster Risk Reduction (DRR) being "an integrated system of hazard monitoring, forecasting and prediction, disaster risk assessment, communication and preparedness activities systems and processes that enables individuals, communities, governments, businesses and others to take timely action to reduce disaster risks in advance of hazardous events" [1]. EWS contribute in reducing vulnerability to floods in urban and rural areas, which can also affect Sustainable Development in the latter. The Sendai framework for Disaster Risk Reduction 2015–2030 recognized seven strategic targets including EWS. The framework also identified four priorities, the fourth of which embeds the concept of Climate Services (CS) as a powerful tool for more effective disaster preparedness and the 'build back better' principle. In this respect, the European research and innovation roadmap for Climate Services expands the contribution of CS, particularly "hydrometeorological services", to the Sendai framework and EWS, a building block of preparedness [2].

The scientific literature presents two main approaches for EWS: top-down and hazard-centered and bottom-up people-centered [3,4]. Kelman and Glantz [5] also categorized EWS by "First Mile" and "Last Mile" approaches, where the former involves communities from the beginning and the latter concentrates on technical solutions and include communities only at the end of the process. As is currently widely acknowledged, top-down or last-mile approaches concentrating on developing forecasts, methodologies and models show many limits in effectively reaching and engaging communities [6,7].

Experiences of bottom-up flood EWS can be retrieved from South Asia, particularly Nepal, India and Bangladesh, all of which practice flood EWS on different scales and use different approaches. Some trans-border flood EWS experience also exists between those countries [8]. In Asia, a strong emphasis on people-centered EWS was observed after the Third International Conference on EWS in Bonn in 2006 and is described as "a complete set of components that connects those who need to receive messages from others who compile and track the hazard information of which messages are composed" [9].

A new paradigm shift was achieved with the concept of Community-Based Early Warning Systems (CBEWS) supported by international non-governmental organizations. A CBEWS is "an early warning system (EWS) where communities are active participants in the design, monitoring and management of the EWS, not just passive recipients of warnings" [10]. The implementation of CBEWS is documented by many authors in developing countries, such as Malawi [11], Nepal [10,12], Indonesia [13], India [14] and Cambodia [15]. CBEWS have proved successful in saving lives, but are often limited in their forecasting time, reducing their suitability for saving assets or livelihoods [16]. Top-down and bottom-up approaches, therefore, need to be integrated: people-centered approaches have to be coupled with robust flood monitoring and forecasting systems.

Since 2015, and according to the recommendation of the World Meteorological Organization (WMO) [17], the need has arisen for EWS to be built on impact-based forecasting and warning services. The aim is to bridge the gap between hydrometeorological forecasts and the potential consequences of the forecasted hazard on specific sectors. This approach, linking forecast information to decision-relevant impact thresholds for users, improves uptake and effectiveness [18]. Although only recently introduced, best practices can be found in the national meteorological services of developed countries, such as the United Kingdom Met Office and National Weather Service of the United States, and are being tested in South Asia [19].

In West Africa, EWS have been conceived and implemented since the 1980s in the sector of food security, mainly addressing agricultural drought [20]. However, efforts surrounding flood management in West Africa have, for the most part, focused on rescue and relief during and after events, while scientific and technical attempts to simulate runoff and forecast flood behavior are limited due to the poor gauging of rainfall and discharge. In recent years, some international initiatives have addressed flood forecasting at global [21] continental [22] or river basin [23] levels to respond to the growing need for flood risk early warning. Despite this effort, none of the web-based systems that are used for ongoing transnational flood forecasting are connected to local EWS, even if they can provide valuable inputs for them. In West Africa, and probably across the whole continent, CBEWS for floods conceived through impact-based forecasts and warnings are not yet documented by the scientific or gray literature.

The objective of our research was to demonstrate that it is possible to set up a comprehensive Community and Impact Based EWS (CIBEWS), responding to the key points and indicators described in the literature, by enhancing existing tools, experience and knowledge in a remote rural area of a poor developing country, such as Niger. This paper describes the approach adopted by the Niger government, with the support of other technical partners, in the setting-up of a CIBEWS in Niger on a Sahelian tributary of the Niger river: the Sirba river. The Local Floods Early Warning System for Sirba, called SLAPIS (Système Locale d'Alerte Precoce contre les Inondations de la Sirba), has been set up within the ANADIA2 (Adaptation to Climate Change and Disaster Risk Reduction for Food Security—Phase 2) project by the National Directorate for Hydrology (DH, Niger) in collaboration with the National Directorate for Meteorology (DMN, Niger), the Interuniversity Department of Regional and Urban Studies and Planning (DIST) Politecnico and University of Turin (Italy) and the Institute for the BioEconomy of the National Research Council (IBE-CNR, Italy). The project was funded by the Italian Agency for Development Cooperation (AICS). The advantages of attaining the SLAPIS objectives are reducing the impacts of floods in both rural (contributing to Sustainable Development) and urban areas, namely that of Niamey, which is a few kilometers downstream of the Sirba–Niger confluence.

#### **2. Materials and Methods**

#### *2.1. Study Area and the Hydrological Context*

The area of interest is located in the Sirba river basin, the main tributary of the Niger river in the Middle Niger River Basin (MNRB). The Sirba river basin covers an area of approximately 39,000 km2 across Burkina Faso (93% of the basin) and Niger in the central Sahel. The territory has a granitic substrate and a slight height variation between the upper level of 444 m a.s.l. and the lower of 181 m a.s.l.. The climate is semi-arid with a long dry season and a rainy season concentrated in 3 to 4 months, between June and September, and an annual rainfall between 400 and 700 mm [24]. The Sahelian climate is characterized by strong rainfall variability with persistent dry spells and extreme rainfall events [25]. Therefore, the hydrology of the Sirba river is determined by the monsoon season and its spatio-temporal variability. The flood magnitude is more influenced by surface runoff than by groundwater flow [26].

The Nigerien sector of the Sirba river was chosen as the study area. The reach covers 108 km, from the state border with Burkina Faso, a few kilometers downstream of the confluence of the three main tributaries (Yali, Faga and Koulouko rivers), to the confluence with the Niger river (Figure 1). According to the last census (2012), the Nigerien part of the Sirba basin has 171 villages with a total population of 88,863. The majority (97 settlements) are distributed in riverine areas, meaning that 61,703 people, belonging to 7732 households, live in potentially flood-prone zones.

**Figure 1.** The geographical framework of the study area [27]. Bossey Bangou hydrometer (BB); Garbey Kourou hydrometer (GK).

Since the beginning of the century, as first highlighted by Tharule [28], extreme floods have been a crucial issue in the development of Sahelian countries. Indeed, an increasing number of flood events and flood-related impacts has been reported by many authors [29] and the frequency of flood events is particularly alarming in the MNRB [30–32]. These events often have disastrous consequences for the population, infrastructure, environment and economic sectors. During the last decade, the floods that struck Ouagadougou and Bobo Dioulasso (Burkina Faso) in 2009, the series of floods that hit Niamey (Niger) in 2010, 2012, 2017 and 2019 and those affecting Mopti and Bamako (Mali) in 2019 were particularly significant.

The high occurrence of these catastrophic events, despite the limited recovery of the climatic trends from the long drought that affected the region in the 1970s and 1980s, is referred to as the "Sahel Paradox" [33,34]. Hydrological studies conducted over the last 25 years clearly show two opposing phenomena: a runoff reduction in Sudano–Guinean catchments and an increase in Sahelian catchments [35–37]. Many researchers claim that, in the Sahel, besides the recent recovery of rainfall, which is still below the pre-1970 levels, and the increasing occurrence of extreme rainfall events, the main driver of floods is the strong land/vegetation degradation that has progressively reduced the water-holding capacity of the soil, leading to greater and faster runoff [33,38,39]. Indeed, even in the context of a so-called "regreening" of the Sahel, the recent increase in seasonal greenness at the Sahelian regional scale [40], investigations have highlighted that this vegetation evolution is not spatially uniform, and large areas remain affected by degradation, such as the northeast of Burkina Faso and the southwest of Niger.

The joint impact of land degradation and extreme rainfall increase produced an extension of the drainage network and the rupture of endorheic basins that caused a further discharge increase [39]. The right bank tributaries of the Niger river, and among these the Sirba river, show discharges 150% higher and runoff coefficients three times higher than those observed up to 50 years ago [31].

In Niger, the increase in flood events has been demonstrated to be country-wide by Fiorillo et al. [41], analyzing official data collected by the government on damages from 1998 to 2017. Regarding the regional and sub-regional impacts of floods, the southwestern areas of the country were confirmed to be the most exposed to flood risk. Over the last 20 years, the scientific literature has focused mainly on changes in Niger river flood magnitudes, trying to understand both the changes underway in regional hydrological characteristics and the main factors triggering the increase in floods in the area. However, Tiepolo et al. [42,43] demonstrated that the Niger river is just one of the causes of the flood risk, with other mechanisms and triggers being present.

#### *2.2. Methods for System Set-Up*

According to the United Nations International Strategy for Disasters Reduction (UNISDR), the four pillars of EWS, SLAPIS has been set up through a progressive process (Figure 2), addressing (1) risk knowledge, (2) risk monitoring and warning, (3) risk information dissemination and communication, and (4) the response capacity of communities and the authorities to respond to the risk information. Approaches, methods, data collected and analysis are described herein, according to each of the four pillars.

**Figure 2.** Conceptual framework of Système Locale d'Alerte Precoce contre les Inondations de la Sirba (SLAPIS) organized on the four pillars of EWS (United Nations International Strategy for Disasters Reduction).

#### 2.2.1. Risk Knowledge: Risk Assessment at Local Level and Flood Scenarios

The risk assessment activities have been performed in the four main rural communities distributed along the last 40 km of the Sirba river: Tallé (population 2603 in 2012), Garbey Kourou (4634), Larba Birno (4713) and Touré (4065). The methodology adopted, as described by Tiepolo et al. [44], considers the risk (R) as "the probability of occurrence of hazardous events or trends multiplied by the consequences if these events occur" [45]. The potential damages have already been used as a component of the risk function in both Niger [46] and the developing countries of South Asia.

Risk assessment is conceived as a process starting from the understanding of the local risk governance framework including local planning processes, national guidelines and the literature. The second step is the identification, through meetings with each community, of the hydro-climatic threats, past catastrophic events, rainfall threshold, and flood level above which damages are registered and local resources mobilized by each community (capacity and assets) to address them. The last phase is the calculation of flood probabilities and the establishment of flood scenarios. Flood scenarios were calculated through the development of an ad-hoc hydraulic numerical model simulating the river behavior for each discharge threshold in the Hydrologic Engineering Center's River Analysis System (HEC-RAS) [47] environment. The hydraulic model was implemented on a topography based on a 10 m Digital Terrain Model (DTM) detailed with Geographical Positioning System (GPS) topographical surveys and calibrated with discharge and level observations, as described by Massazza et al. [27]. In order to take into account changes in the hydrological behavior of the Sirba river over time, as described by Tamagnone et al. [26], non-stationary analyses were conducted to identify the probability of occurrence of the assigned hazard thresholds [27,48,49]. The hazard thresholds, as already described by Massazza et al. [27], are shown in Table 1.

**Table 1.** Hazard thresholds for different flow conditions of the Sirba river [27]. Discharge (Q); Flow Duration Curves (FDC); Stationary Generalized Extreme Value (SGEV); Stationary Return Period (S-RT); Non-Stationary Generalized Extreme Value (NSGEV); Non-Stationary Return Period (NS-RT).


Each threshold was simulated using the hydraulic model in order to define the area and relative hazard level to which riverine populations are subjected. Hydrological thresholds of flood scenarios were linked to field impacts, according to the national flood hazard classification [50] and international guidelines, such as the WMO Guidelines on Multi-Hazard Impact-Based Forecast and Warning Services [17]. Lastly, the risk level characterizing each scenario was obtained by matching together the information regarding the extent of flood-prone zones, the identification of exposed assets and their value [44].

2.2.2. Monitoring and Warning Service: Hydrological Observations and Forecasts, Data and Information Management

The monitoring component of the system relies on two automatic gauging stations at Bossey Bangou (upstream, at the Burkina Faso border) and Garbey Kourou (downstream, near the confluence with the Niger river). The Garbey Kourou hydrometer was installed in 1956 and is equipped with two water pressure measuring devices, one controlled by the Niger Basin Authority (NBA) and the other by the DH of the Republic of Niger, while the Bossey Bangou gauging station was installed in June 2018, in the framework of SLAPIS, and is managed by DH. The Bossey Bangou (2018–2019) and Garbey Kourou (1956–2019) updated discharge series and a set of 14 discharge measurements were used for hydrological and hydraulic modelling [26]. The headwater of the Sirba river in Burkina Faso is equipped only with hydrometric stations that are non-operational for the real-time monitoring of discharge and, therefore, are useless for the EWS.

Further information on water depth, maximum water levels and flooding extent were collected from local observations and field surveys made at the main localities along the Sirba river. Colored hydrometric staffs (ladders) were installed in May 2019 in five villages along the Sirba river: Touré, Larba Birno, Garbey Kourou and Tallé in the municipality of Gotheye and Larba Toulombo in the municipality of Namaro. The staffs are marked with the four different colored flood scenarios (green, yellow, orange and red). The levels of the colored staffs were defined thanks to fixed topographical points identified during the land surveys. A volunteer observer was appointed within the Community Early Warning and Response System (SCAP-RU) of the village and was trained.

Concerning forecasts, the system relies on two types: hydraulic model forecasts (related to observations of upstream hydrometric stations) and hydrological model predictions (derived from hydrological models acting on the Sirba basin). The first consists of the warning that should be conveyed to villages in the case of the river passing the hazard threshold at the upstream hydrometer. The hydraulic model allowed the flood propagation time to be calculated and, thus, the warning time for each village [44]. This type of forecast has a higher level of certainty but may give only a few hours or up to one day of notice to the riparian villages downstream.

Hydrological model forecasts have a major uncertainty but can give indications towards the evolution of the hydrology up to 10 days in advance. At present, the early warning system bases its forecasts on the global hydrological model GloFAS 2 [21]. Preliminary analysis shows that the gap between observed and forecasted discharge is quite significant. This suggested the post-processing of forecasts in order to decrease the bias and improve the EWS reliability. GloFAS forecasts are adjusted with corrective factors, improving their reliability according to historical series and real-time measured data. The optimization process was conducted through the linear regression method over homogeneous periods of the rainy season and was based on 10 years of simulations (2008–2018). The optimization allows quality improvement with an increase in Root Mean Square Error (RMSE) and the Probability of Detection (POD) of extreme events and, at the same time, reduces the False Alarm Rate (FAR), as described by Passerotti et al. [51]. A further improvement in the forecasting system is foreseen with the integration of a second model, Niger-HYPE [23].

Data management and services are ensured by a Spatial Data Infrastructure (SDI) based on interoperable and open source solutions and Open Geospatial Consortium (OGC) web services [52] for the management of observed and forecasted data and the establishment of a hydrological warning communication service. Methodologically, the implementation steps were the conceptual and formal data model design, the development of the SDI, the setting up of some Open Web Services (OWS) standards through the development of services and procedures for data flow management, the forecast data optimization and geoprocessing functions.

The SLAPIS client–server architecture (Figure 3) is based on open source technologies and software components which allow it to interact between data providers and end-users, including three main layers: data retrieval and storage, web services and user interface. All data are managed by a central open source geodatabase, which is the core of the SLAPIS server. Geoprocessing routines and data optimization procedures have been implemented on the data layer in order to ensure that the observed and forecasted data are uploaded into the system data model. Furthermore, Application Programming Interfaces (APIs) have been developed both to transfer forecast data from Niger-HYPE and GloFAS platforms and to foster the communication among the SDI components. For retrieving data from providers not equipped with standard and interoperable web services, we used the File Transfer Protocol (FTP).

For the front-end of the system, a customized Graphical User Interface (GUI) was designed and implemented for monitoring, in quasi-real-time, the observed and forecasted data and their visualization in graphic and tabular formats. The customized functions allow the users to retrieve (from the GUI) the entire data set for further analysis or applications. SLAPIS also has an open data portal which, using the Comprehensive Knowledge Archive Network (CKAN) [53] open source data catalogue, allows access to the available data, including raw and intermediate research data, as well as complementary studies on the area. Each dataset recorded in CKAN contains a description of the data and other useful information, such as available formats, the producer (if they are freely available) and the topic. Finally, a simple information box is available on the main page and it is automatically updated by the system with the current state of vigilance.

**Figure 3.** The SLAPIS information system architecture.

#### 2.2.3. Dissemination and Communication: Stakeholders' Consultation

The dissemination and communication mechanism of SLAPIS was defined by a stakeholders' consultation and an analysis of the national alert mechanism. As indicated by the National Alert Code [54], the warning measures are disseminated by the decision of the Minister of Civil Protection at a national level, by Governors in the regions, Prefects in departments and Mayors in the municipalities (Article 5). Alert messages are prepared from information provided by technical institutions at different administrative levels.

An analysis of the needs of the actors in terms of information on the flood risk was performed through semi-structured interviews with national stakeholders, technical workshops with local administrations and focus groups with the communities involved. The result was the definition of the SLAPIS communication and information plan, stating that information produced by SLAPIS should be accessible to all stakeholders through specific tools and channels. Subsequently, information on the state of vigilance is transformed into alerts by the competent institutions according to the magnitude and amplitude of the forecasted flood risk.

Concerning the last-mile of communication, challenges include access to the information, the ability to understand the warning, and the ability to take action [55,56]. All these issues were dealt with via focus groups with local governments and community representatives. A set of actions were defined in order to create awareness at a community level about the flood scenarios and the actions to be taken in the case of a warning. Among different approaches, visualization is the one that has been preferred to aid the interpretation of flood scenarios [57]. In the context of SLAPIS, visualization includes the adoption of the four-color classification for scenarios. They are associated with warning content (the core of the message is the color), with the colored hydrometric staff gauges (qualitative gauging staffs) installed along the river, as well as information panels in the villages indicating priority actions to be taken. We adopted the four color-coded classes currently used by different countries (i.e., United Kingdom—the Met Office; EU—Meteoalarm, Philippine—Pegasa, Italy—Protezione Civile, India Meteorological Department) [19] and responding to the international standards (International Organization for Standardization - ISO 22324:2015). The classes are related to discharge, return periods and impacts on the main riverine settlements according to the classification, as described by Massazza et al. [27]—essentially, green stands for the normal condition, meaning a no-impact scenario, yellow (Stationary Return Period 10 years) stands for minor impacts, orange (Stationary Return Period 30

years) stands for significant impacts and red (Stationary Return Period 100 years) stands for severe impacts. Moreover, the installation of colored hydrometric staffs aims to increase awareness of the flood risk among communities by showing the levels of the hazard thresholds—the height that the flood can reach. Local hydrometric staffs also aim to establish a local communication system, building on the approach described by many authors in Asia [15,58], between upstream and downstream villages.

#### 2.2.4. Response Capability: Communities Preparedness and Action

According to Girons Lopez [59], response capability was based on social preparedness for flood loss mitigation. Community flood risk reduction plans were prepared for the four main villages of Touré, Larba Birno, Garbey Kourou and Tallé in the municipality of Gotheye. The plans have the objective of associating the flood scenarios and the stakes to underline the specific criticalities of each village and propose measures to reduce potential damage. As described by Tiepolo et al. [44], the plans were drawn up with a multi-step methodology: participatory hazard identification, probability of flood occurrence, flood-prone areas, asset (mostly housing and crops) identification and risk reduction actions. The assets are identified in the flood zone by municipal technicians integrated using very high photointerpretation, as described by Belcore et al. [60]. Actions include both risk prevention and the preparedness actions known by the target communities, as well as best practices from the reference literature.

According to Fakhruddin et al. [61], the participatory development of flood risk reduction plans has the objective of enabling people to act, empowering communities with basic knowledge of the flood risks and of more urgent actions to be taken according to each scenario. Actions to reduce risk are associated with the flood scenario and the four color-coded classes of warnings. Therefore, the warnings embed both physical information (water depth and flood zones) and social information (such as community assets likely to be affected and community actions to be taken).

Community preparedness has also been strengthened by adopting and adapting the approach developed by Stitger et al. [62,63] for drought risk management through Roving Seminars on agrometeorology and agroclimatology. A new concept of Roving Seminars for flood risk management has been developed. The seminars take the form of a one-day meeting in a village, which the whole community is invited to attend. The objective is to make communities become more self-reliant in dealing with hydrometeorological issues related to floods that affect human life, habitats, assets, livestock and crops, and to increase the interaction between the community and the National Meteorological and Hydrological Services.

#### **3. Results**

The results are reported in the following sections, relating to each of the four pillars of EWS.

#### *3.1. Risk Knowledge*

The first main result in the definition of risk level was the assessment of flood scenarios. They were defined on the hazard thresholds, fixed on both the statistical analysis of discharge and impacts on the main riverine settlements according to the four color-coded classes. Flood hazard scenarios were mapped, showing the extension of flood-prone areas (Figure 4). The bulk of the assets are located on the left bank of the Sirba river (houses, community services, infrastructure, fields and vegetable gardens) while the assets on the right bank are essentially fields and orchards with few settlements.

The hydraulic numerical model was also used to calculate the conveyance time of the flood wave: the upstream hydrometer of Bossey Bangou provides notice of between 20 (Touré), 26 (Larba Birno) and 28 hours (Garbey Kourou and Tallé) [27]. Scenarios include the identification of exposed assets (houses, orchards, crops, pits, barns and wells) and their value, as described by Tiepolo et al [46]. Table 2 reports the value of the assets that could be damaged in the four main riverine villages by a flood event with a magnitude equal to the hazard threshold. Reported amounts should be considered,

keeping in mind that the average annual GDP per capita in Niger is 430€ (2018) and the minimum wage is 46€ per month.

**Figure 4.** Atlas of flood-prone areas of the Sirba river. Tallé village (Gotheye municipality, Niger). This image reproduces the Table C1 of the Atlas which is in the official language of Niger (French). The Atlas is annexed as a Supplementary Material.

**Table 2.** Cumulative value (k€) of exposed assets in the four main villages along the Sirba river (adapted from Tiepolo et al. [44]).


#### *3.2. Monitoring and Warning Service*

The observed and forecasted data are accessible by stakeholders using the SLAPIS web platform (www.slapis-niger.org), with specific characteristics that make it unique in the panorama of web tools developed for alerting at local level, because it integrates the following five levels (Figure 5):


**Figure 5.** Structure of the SLAPIS monitoring and warning services.

The observed data are also accessible through the CKAN catalogue (http://sdicatalog.fi.ibimet.cnr.it: 5003/fr/dataset?groups=slapis\_prj), as well as other geographical data used by the system in different formats (SHP, GeoJSON, JSON and CSV).

The network of local observers is composed of two at the gauging stations and five village observers at the colored hydrometric staffs (Figure 6).

**Figure 6.** Map of the observation network and the flood propagation time.

In order to increase awareness of the flood risk among communities and integrate the observation network with additional local measures, colored hydrometric staffs (Figure 7) have been installed in five villages along the Sirba river with the objective of showing the levels of the hazard thresholds, the height that the flood can reach, to increase awareness of the danger of flooding and communicate the level of risk to villages downstream. The observers have been trained to read the scales and communicate any rise in the waters to the DH, the Vulnerability Monitoring Observatory (OSV) located

in the municipality and the SCAP-RUs of the villages downstream, according to a specific observing protocol. In 2019, they used a WhatsApp group to send the information as a photo; however, in 2020, a system of visualization of the observed water levels at the colored staffs will be integrated into the SLAPIS platform.

**Figure 7.** Colored hydrometric staff at Garbey Kourou, Niger.

#### *3.3. Dissemination and Communication*

SLAPIS is integrated in the national alert system; indeed, its information mechanism was defined thanks to an analysis of the national alert mechanism and an analysis of the needs of the actors in terms of information on the flood risk. The communication and dissemination plan of SLAPIS defined that the flood scenarios produced in the framework of SLAPIS and the related warnings on the level of vigilance are accessible to all stakeholders with specific tools. Subsequently, the state of vigilance is transformed into an alert by the competent institutions. Figure 8 summarizes the SLAPIS information mechanism. In particular, it indicates the actors to whom the information is communicated directly, through which channel and in what format.

The information on the state of vigilance is then transformed into an alert by the competent institutions according to their protocol and communicated through institutional channels, as recommended by Rahman et al. [64]. According to Oktari et al. [65], the communication system is multi-channeled in order to ensure maximum outreach. Specific communication channels are established with different types of stakeholders. There are four main official information flows:

	- o The OSV through an emergency meeting. The OSV alerts sectoral infrastructure (schools, water supply, etc.);
	- o Community Radio by telephone to dictate an alert release;
	- o Chiefs of concerned villages by telephone. The village chief mobilizes all available means to alert the population (loudspeaker, town crier, etc.);

In addition to the information from the information system, SLAPIS integrates local observations made at the colored staffs installed in the main villages bordering the Sirba. The observer appointed within the SCAP-RU is responsible for communicating the possible rise in the water level to the OSV located in the municipality and to downstream villages using the color code of the flood scenarios (green, yellow, orange and red).

**Figure 8.** Information flow (Hydrology Directorate (DH); Ministry of Humanitarian Action and Disaster Management (MAH); Regional Committee of Food Crisis Prevention and Management System (CRPGCCA); Departmental Committee of Food Crisis Prevention and Management System (CSRPGCA); Vulnerability Monitoring Observatory (OSV); General Directorate for Civil Protection (DGPC); Regional Directorate for Civil Protection (DRPC); Department Directorate for Civil Protection (DDPC)).

#### *3.4. Response Capability: Communities Preparedness and Action*

Since 2006, the WMO has been promoting the organization of one-day Roving Seminars on weather and climate for farmers. We adapted the concept of Roving Seminars to the hydrological risks, with the main objective of strengthening communities' self-reliance in dealing with floods and other extreme hydrometeorological risks. Moreover, the seminars increase the interaction between the local communities and local staff of the National Meteorological and Hydrological Services, building trust and confidence. The seminars are organized in two parts. The first focuses on the weather, climate and hydrology of the region, particularly on the relationship between rainfall and discharge on the whole river basin and on local sub-basins. The aspects of change in land-use/land-cover and changes

in hydrological processes are also well developed in order to promote a better comprehension of the dynamics, in which farmers are a key actor. The second part of the day focuses on the communities' perception of weather and hydrological information/alerts provided by SLAPIS. The main objective is to obtain feedback from the communities by a free and frank exchange of ideas and information. This part of the seminar is designed to engage all the participants in discussions and obtain suggestions on additional information and ways and means of improving future communication to facilitate effective operational decision making.

SCAP-RU are key actors in monitoring and coordinating actions in an emergency and in peacetime. SLAPIS invested in the empowerment of SCAP-RU, both technically and socially. From a technical point of view, SCAP-RU have been trained in risk assessment, monitoring and impact observation. Each SCAP-RU has been equipped with a smartphone and access to the internet to collect and transmit observations and receive information and alerts. SCAP-RU members have been charged with specific roles before, during and after emergencies, reinforcing their status and becoming a reference within the community.

In Niger, local plans to address hydro-climatic hazards are not required by law. Within SLAPIS, we developed local flood risk reduction plans for the main villages exposed to flood risk. The plans were coordinated with the communities and organized considering local capacities and assets. Even if the plans are very simple, they include a contingency part involving the setting up of an emergency committee, a map of the areas according to flooding probability and a map of exposed assets and basic emergency instructions. The plans also include structural prevention measures, such as the protection or displacement of boreholes, wells, fountains and photovoltaic plants [44]. The visualization of contingency actions in relation to risk levels was realized through local information panels, which report the main actions to be taken by communities in text and graphics.

Another result was the production of a Cartographic Atlas of flood-prone areas (published as a Supplementary Material to this paper) over the 108 km of the Nigerien reach of the Sirba river. Maps are compiled at large (1: 100,000) and detailed (1: 20,000) scales. The most exposed riverine villages of Tallé, Garbey Kourou, Larba Birno, Larba Toulombo, Guidare, Toure, Boulkagou and Bossey Bangou are also represented at greater detail (1: 5000).

#### **4. Discussion**

The results of the study are discussed herein according to the indicators identified by Sai et al. [19] for steering each of the four components of effective EWS (See Table A1 in Appendix A).

Concerning risk knowledge, the indicators identified in the literature are:


Concerning the monitoring and warning components, the indicators proposed by the literature are:

• Timely and accurate forecast—good quality data have to be collected and processed in real or quasi-real-time to produce meaningful, timely and accurate forecasts [3]: hydrological forecasts

can ensure a lead time of 10 days, but their accuracy is lower than hydraulic forecasts based on the upstream hydrometer [27], which have a lead time of 28 hours. The accuracy of hydrological forecasts has been improved, assimilating real-time data observed at the gauging stations at Bossey Bangou and Garbey Kourou;


Concerning dissemination and communication, the indicators collected by Sai et al. [19] are:


and actions. Each SCAP-RU has knowledge of the scenarios associated with alert colors, flooding area maps, assets at risk and priority actions to be taken. Participatory meetings proved to be really useful in building trust.

Concerning response capability, the indicators addressed are:


Finally, some cross-cutting indicators are also addressed:

• Local community participation—end users can actively contribute to all four components of EWS [3,61,64,71]: communities are key actors in Participatory Local Risk assessment [44], discharge monitoring at a local level (SCAP-RU), communicating the observed levels of vigilance (SCAP-RU) and defining the contingency actions of risk reduction [46]. Furthermore, the technical architecture and GUI of the SLAPIS are designed to integrate (i) top-down and bottom-up approaches, (ii) hydrological forecasts and observations with local perception of populations and a lead time suitable for operational decision-making processes at national and local levels. According to many authors, involving communities and local stakeholders is the main challenge for achieving the purpose of the EWS and SLAPIS proved that it can be effectively achieved by appointing local observers, organizing meetings with communities, involving them in the risk knowledge phase and jointly defining communication and risk reduction plans.

#### **5. Conclusions**

According to Cools et al. [71], the key point to enhance the effectiveness of a flood EWS is: "a better match between the available risk information, the forecasting system and the response capability of authorities and the at-risk population". Cools et al. suggested, and the present study demonstrates, that "Engaging local communities and authorities in the EWS design can improve the effectiveness of the whole early warning process and hence results in a higher response to an alert warning".

The study proves that such key points can be operationally addressed, leveraging existing resources, local stakeholders and knowledge using simple but effective approaches and integrating state-of-the-art hydrologic-hydraulic scientific results in a decisional scheme for Sahelian rural areas. This mechanism will be replicable in each context, even if characterized by knowledge and structural deficits, creating a better capacity to exchange data and information and by directly connecting available technical capabilities with the local level. Beneficiaries are, therefore, able to define the rules that will develop the whole system, which, in any case, needs to be consistent with the legislation in force in the country and with internationally recognized best practices.

This study suggests that, instead of developing new forecasting tools, it can be preferable to enhance those already operating on the basin and calibrate them on the local scale by adding real-time observation control points and to connect discharge thresholds, field observations and hydrological forecasts with potential impacts through flood scenarios. This multidisciplinary approach ensures a greater level of suitability and sustainability. Indeed, it allows us to enhance the resonance of hydrological models already developed by the scientific community, which are not usually exploited by local technical structures and to concentrate efforts on: (1) the downscaling of forecasts, (2) topographical surveys and discharge measurements on-site, (3) the quality improvement of observations and (4) the implementation of hydraulic models to guide the planning of mitigation and adaptation strategies. The strength of simplicity also lies in not having to spread complex messages, but simply the reference risk scenario and, finally, its color-code (according to the international standards of ISO 22324:2015), which already embeds all of the other information.

The main limit of the study is that it focuses only on the Sirba reach in Niger. The absence of trans-border flood risk assessment and the lack of real-time hydrological monitoring upstream in Burkina Faso prevent the flood risk information being spatially extended upstream. Moreover, the limited lead time provided by the observation at the Bossey Bangou gauging station to the main downstream villages is a limitation of the EWS on the Sirba river.

The ongoing improvements of SLAPIS, already planned before the 2020 rainy season, include implementation of: (1) a second optimization based on real-time observed discharge and (2) the operational integration of the forecasting system of the regional hydrological model Niger-HYPE, ensuring more resilience and more accurate discharge forecast in the system. Further future developments of this study include the improvement and extension of the existing flood EWS to the whole Sirba basin. Naturally, this implies an urgent need for systematic flood monitoring and communication between the two countries (Burkina Faso and Niger), as well as a coherent flood risk assessment performed upstream of the boundary. In this respect, the respective governments' commitment to sharing of information and effectively disseminating it to flood-prone communities is necessary. At broader level, the study found that there is a need for institutionalizing and strengthening the existing practices of sharing of flood information between different countries for EWS purposes. The simple and integrated approach illustrated in this case study, bridging the gap between top-down and bottom-up approaches described by the literature, can inspire governments, local administrations and development partners to invest in the improvement of existing tools and knowledge in order to strengthen cooperation, collaboration and coordination, reduce hazard impacts and sustain the development of rural and urban areas.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2071-1050/12/5/1802/s1, Atlas of flood-prone areas of the Sirba river (Atlas cartographique des zones inondables de la Rivière Sirba.

**Author Contributions:** Conceptualization: V.T.; Methodology: V.T., M.T., M.R., A.P., T.D.F. G.L.K.; Developing: L.R., T.D.F., E.R., G.M., P.T.; Validation: L.R., E.R., G.M. P.T.; Investigation: G.M., P.T., M.H.I., V.M.; Writing—original draft preparation: V.T.; Writing—review and editing, M.T., T.D.F., L.R., V.M., G.M., P.T.; Funding acquisition: V.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was co-funded by the Italian Agency for Development Cooperation, by the Institute of Bioeconomy (IBE)-National Research Council of Italy (CNR), by the National Directorate for Meteorology of Niger (DMN) and by DIST Politecnico and the University of Turin within the project ANADIA2.

**Acknowledgments:** We would like to thank Moussa Labo (former Director of DMN) for trusting in the project. We would like to express our deepest gratitude to the Joint Research Center of European Commission and European Centre for Medium-Range Weather Forecasts (JRC-ECMWF) and the operational flood forecasting and alerts in West Africa (FANFAR) Project for belief in a collaborative approach and providing access to GloFAS and Niger-HYPE forecasts. We thank Bruno Guerzoni, Hamidou Issa Mossi (Mayor of Gotheye), Abdoul Wahab Oumaru (Mayor of Namaro), Ibrahima Djibo (Mayor of Torodi), Souradji Issa (DDA/Gotheye), for support and participation. Finally, the authors want to thank, with affection, Steffen Muller (GIZ), discussions with whom led to the idea of SLAPIS being born in the hot Nigerien evenings.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.


### **Appendix A**



*Sustainability* **2020**, *12*, 1802


*Sustainability* **2020**, *12*, 1802


**TableA1.***Cont.*

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
